CN-122020406-A - Fault diagnosis method for sewage treatment failure

CN122020406ACN 122020406 ACN122020406 ACN 122020406ACN-122020406-A

Abstract

The invention belongs to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method for sewage treatment failure. The method comprises the steps of collecting flow monitoring data of a sewage treatment system, extracting real-time fluctuation and daily trend characteristics through self-adaptive multi-scale attention preprocessing, screening an effective data set of focusing faults, constructing three fault subtrees of association, cross-link coupling and independent type, establishing an association weight matrix between the subtrees to form an association enhanced improved fault tree model, carrying out layered correction on the fault probability of a bottom event and the subtrees, combining global association strength to obtain total fault probability, setting a threshold value to judge faults, and realizing accurate positioning according to the subtree probability and the association weight. The invention improves the calculation precision of the fault probability and realizes accurate fault positioning. The method can accurately identify the core fault source, avoid misjudgment of cascading failure, provide reliable support for intelligent operation and maintenance of the sewage treatment system, and solve the pain point with insufficient diagnosis precision in the traditional method.

Inventors

LI KUNQUAN
ZHANG JIFEI
ZHU MINGXIANG
SUN KAI

Assignees

冠县瑞冠再生资源有限公司

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (6)

1. A fault diagnosis method for a wastewater treatment failure, comprising the steps of: s1, collecting flow monitoring data of a sewage treatment system to form an original data set; S2, performing self-adaptive multi-scale attention preprocessing, constructing a double-scale feature extraction unit and preprocessing of an attention mechanism, and processing an original data set to obtain an effective data set of key features of focusing faults, wherein the double-scale feature extraction unit respectively captures real-time fluctuation features and daily trend features, and the attention mechanism highlights the features with high fault correlation degree through dynamic weight distribution; s3, constructing an associated fault subtree, a cross-link coupling fault subtree and an independent fault subtree according to the effective data set, establishing an associated weight matrix among the subtrees, quantifying the mutual influence degree of fault events of different subtrees, and combining to obtain an associated enhanced improved fault tree model; S4, according to the association enhancement type improved fault tree model, firstly deconstructing top events, subtrees and bottom events of the fault tree model, distinguishing internal associations of the events in the subtrees and external associations among the subtrees; s5, setting a sewage treatment fault judging threshold, judging that no fault exists if the total fault probability is smaller than the judging threshold, judging that a fault exists if the total fault probability is larger than or equal to the judging threshold, and performing fault positioning diagnosis processing when the fault exists.
2. The fault diagnosis method for sewage treatment failure according to claim 1, wherein the step S2 is performed with adaptive multi-scale attention preprocessing, a preprocessing model of a dual-scale feature extraction unit and an attention mechanism is constructed, an original data set is processed, and the specific implementation of obtaining an effective data set of focus fault key features is as follows: Constructing a double-scale feature extraction unit, respectively extracting real-time fluctuation features and daily level trend features, and setting a real-time fluctuation feature extraction sliding window Extracting real-time fluctuation characteristics by adopting a sliding window time domain difference method, wherein the calculation formula is as follows: , wherein, For the feature mean value within the sliding window, For the real-time fluctuation value of the jth monitoring index of the ith sampling point, k is the sample index in the sliding window, t is the real-time fluctuation identifier, and all the real-time fluctuation values form a real-time fluctuation feature set ; Setting a period of aggregation of heliostat-level trend features The daily trend characteristics are extracted by adopting a daily time aggregation weighted average method, and the calculation formula is as follows: wherein In order to time-decay weights, The daily trend value of the jth monitoring index is taken as the ith time sampling point, and all daily trend values form a daily trend feature set ; Calculating mutual information values of the daily trend feature set, the real-time fluctuation feature set and the fault label Y by adopting a normalization mutual information method: , wherein, The mutual information value of the jth daily trend feature or the real-time fluctuation feature and the fault label, ; Generating attention weight coefficients based on the fault association degree, wherein the calculation formula is as follows: Setting attention weight screening threshold values, respectively screening out indexes meeting weight conditions in daily trend and real-time fluctuation feature sets, and taking intersection of the indexes to obtain an effective data set.
3. The method for diagnosing a wastewater treatment failure according to claim 1, wherein the feature cluster classification of the features in the effective dataset is required before constructing the association-enhanced improved fault tree model, and is specifically implemented as follows: Firstly, determining a threshold interval for each feature in the effective data set based on historical normal operation data, and judging that the feature is in an abnormal state when the feature value does not belong to the threshold interval; identifying linkage association degree of characteristic abnormality of subsequent links caused by the preceding flow links through Granges causal value Wherein As a result of the graininess of the cause and effect values, Calculating the correlation degree of the mutual influence of different link characteristics by adopting normalized mutual information values at the same time Calculating the independence of the single feature on the fault label by adopting the variance contribution rate ; Clustering is carried out according to the three linkage association degrees, the mutual influence association degrees and the independence degrees as clustering centers, and three different characteristic clusters are obtained.
4. The method for diagnosing a failure in sewage treatment according to claim 1, wherein the step S3 is characterized in that an associated fault subtree, a cross-link coupling fault subtree and an independent fault subtree are constructed according to an effective data set, an associated weight matrix between the subtrees is established, the degree of interaction of fault events of different subtrees is quantified, and the specific implementation of obtaining an associated enhanced improved fault tree model is as follows: Marking three different feature clusters as three event sets, and constructing three fault subtrees with different structures; Constructing an independent fault subtree by taking a single link failure as a top event and a single characteristic abnormal event in an independent event set as a bottom event, taking a follow-up link cascading failure as a top event, taking a preamble characteristic abnormal event of a related event set as an initial bottom event and a middle event as a fault propagation node, constructing a related fault subtree, and constructing a cross-link coupling fault subtree by taking a multi-link cooperative failure as a top event and taking a multi-link characteristic abnormal event in a cross-link coupling event set as a bottom event; Calculating direct association weights among all fault subtrees based on mutual information, measuring the influence of the subtrees transmitted by intermediate subtrees to calculate indirect association weights among the subtrees, and carrying out weighted summation on the direct association weights and the indirect association weights to obtain total association weights among the subtrees so as to obtain an association weight matrix; And taking top events of three types of fault subtrees as intermediate events of the whole fault tree, determining the connection strength between each subtree according to the association weight matrix, and integrating to form an association enhanced improved fault tree model.
5. The method for diagnosing the failure of sewage treatment according to claim 1, wherein the step S4 is characterized in that the method comprises the steps of firstly deconstructing a top event, a subtree and a bottom event according to an association enhanced improved fault tree model, distinguishing the internal association of the subtree inner bottom event and the external association between the subtrees, correcting the base event basic probability, calculating the subtree fault probability according to the subtree structure, merging the subtree association corrected subtree probability, and finally combining the top event logic and the global association strength to obtain the concrete realization of the total fault probability: Calculating the correlation correction probability of the bottom event, calculating the average value of the correlation intensities of the bottom event and other bottom events in the same subtree, and correcting by combining the historical statistical basic probability of the bottom event to obtain the actual fault probability of the bottom event: , wherein, The base failure probability for the ith bottom event of the qth subtree, The average association degree of the bottom event and other bottom events in the same subtree is obtained; according to the calculation mode of the type adaptation response structural characteristic of the q-th specific subtree, calculating the initial fault probability of the subtree, and for the independent subtree, adopting logic OR gate operation: For the associated subtrees, adopting propagation probability superposition: and (3) performing logical AND gate operation on the cross-link coupling subtrees: , wherein, For the q-th sub-tree, For the propagation intensity inside the subtree, mq is the number of bottom events of the q-th subtree; Based on the association weight matrix, quantifying the interaction among subtrees, and correcting the initial subtree probability to obtain: , wherein, The total number of specific subtrees in the three kinds of class-associated enhancement type fault tree is respectively, To represent the fault impact weights between subtrees for the established associated weight matrix elements, Summing up specific subtrees of all the non-q-th subtrees for quantifying the comprehensive fault influence of other subtrees on the q-th subtree; Combining the logic association of the top event and three subtrees, introducing global association strength, and obtaining total probability by weighted fusion: wherein Q is the total number of all subtrees, And calculating the global association strength by the average value of all the elements of the association weight matrix.
6. The method for diagnosing a failure in sewage treatment according to claim 1, wherein in the step S5, if the total failure probability is greater than or equal to a decision threshold, it is determined that a failure exists, and when the failure exists, a failure positioning diagnosis process is performed, in which a sub-tree type with the highest failure probability is identified from an association enhanced improved failure tree model, a link of occurrence of the failure is confirmed, and a sub-tree with the greatest influence on the global failure is positioned according to an association weight matrix among the sub-trees, thereby realizing the positioning of the failure.

Description

Fault diagnosis method for sewage treatment failure Technical Field The invention belongs to the technical field of fault diagnosis, and particularly relates to a fault diagnosis method for sewage treatment failure. Background The sewage treatment system comprises a plurality of flow links such as water inlet, biochemical reaction, precipitation, filtration and water outlet, and the like, and monitoring data of each link shows the characteristics of multi-dimensionality and time sequence. In the running process of the system, the fault causes have the characteristics of complexity and concealment, and single-ring-section abnormality can cause multi-ring-section linkage failure. The traditional fault diagnosis method is mostly based on single-scale monitoring data for analysis, and the coupling influence among subtrees is not integrated when the fault probability is calculated, so that the total fault probability calculation accuracy is insufficient, and the actual requirement of efficient fault diagnosis of a sewage treatment system is difficult to meet. Disclosure of Invention The invention provides a fault diagnosis method for sewage treatment failure aiming at the technical problems in the background technology. In order to achieve the above purpose, the technical scheme adopted by the invention comprises the following steps: s1, collecting flow monitoring data of a sewage treatment system to form an original data set; S2, performing self-adaptive multi-scale attention preprocessing, constructing a double-scale feature extraction unit and preprocessing of an attention mechanism, and processing an original data set to obtain an effective data set of key features of focusing faults, wherein the double-scale feature extraction unit respectively captures real-time fluctuation features and daily trend features, and the attention mechanism highlights the features with high fault correlation degree through dynamic weight distribution; s3, constructing an associated fault subtree, a cross-link coupling fault subtree and an independent fault subtree according to the effective data set, establishing an associated weight matrix among the subtrees, quantifying the mutual influence degree of fault events of different subtrees, and combining to obtain an associated enhanced improved fault tree model; S4, according to the association enhancement type improved fault tree model, firstly deconstructing top events, subtrees and bottom events of the fault tree model, distinguishing internal associations of the events in the subtrees and external associations among the subtrees; s5, setting a sewage treatment fault judging threshold, judging that no fault exists if the total fault probability is smaller than the judging threshold, judging that a fault exists if the total fault probability is larger than or equal to the judging threshold, and performing fault positioning diagnosis processing when the fault exists. Preferably, the step S2 performs adaptive multi-scale attention preprocessing, builds a preprocessing model of a dual-scale feature extraction unit and an attention mechanism, processes an original dataset, and obtains a valid dataset of a focus fault key feature, which is specifically implemented as follows: Constructing a double-scale feature extraction unit, respectively extracting real-time fluctuation features and daily level trend features, and setting a real-time fluctuation feature extraction sliding window Extracting real-time fluctuation characteristics by adopting a sliding window time domain difference method, wherein the calculation formula is as follows: , wherein, For the feature mean value within the sliding window,For the real-time fluctuation value of the jth monitoring index of the ith sampling point, k is the sample index in the sliding window, t is the real-time fluctuation identifier, and all the real-time fluctuation values form a real-time fluctuation feature set; Setting a period of aggregation of heliostat-level trend featuresThe daily trend characteristics are extracted by adopting a daily time aggregation weighted average method, and the calculation formula is as follows: wherein In order to time-decay weights,The daily trend value of the jth monitoring index is taken as the ith time sampling point, and all daily trend values form a daily trend feature set; Calculating mutual information values of the daily trend feature set, the real-time fluctuation feature set and the fault label Y by adopting a normalization mutual information method: , wherein, The mutual information value of the jth daily trend feature or the real-time fluctuation feature and the fault label,; Generating attention weight coefficients based on the fault association degree, wherein the calculation formula is as follows: Setting attention weight screening threshold values, respectively screening out indexes meeting weight conditions in daily trend and real-time fluctuation feature sets, and taking intersection of the i