Sewage fault diagnosis method and device based on adaptive dynamic slow feature analysis

Abstract

The invention provides a sewage fault diagnosis method and device based on adaptive dynamic slow feature analysis, and the method comprises the steps: carrying out the classification of online obtained sewage treatment data based on an affinity propagation clustering algorithm, and obtaining multi-cluster online classification data; inputting the multi-cluster online classification data into a pre-constructed slow feature analysis model to obtain online advantage slow features and online disadvantage slow features output by the slow feature analysis model; according to the on-line advantage slow features and the on-line disadvantage slow features, determining corresponding on-line advantage statistics and on-line disadvantage statistics respectively; and comparing the online advantage statistic with the advantage self-adaptive control limit, comparing the online disadvantage statistic with the disadvantage self-adaptive control limit, and determining that the system has a fault based on the fact that the corresponding statistic exceeds the corresponding self-adaptive control limit, otherwise, determining that the running state of the sewage treatment process is normal and no fault exists. According to the invention, timely and accurate fault detection is realized, and loss caused by faults is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a sewage fault diagnosis method and device based on adaptive dynamic slow feature analysis. Background Art

[0002] At present, sewage treatment plants are gradually moving towards automation and intelligent control. With the popularization and application of industrial Internet of Things and intelligent sensor technology, sewage treatment plants can collect a large amount of sewage treatment process data. These data often contain important operation information of sewage treatment plants. With the help of big data analysis, artificial intelligence and other technologies, data-driven sewage treatment process monitoring and fault diagnosis have become a research hotspot.

[0003] Commonly used statistical-based fault diagnosis methods for sewage treatment processes mainly include: principal component analysis, canonical correlation analysis, independent component analysis, etc. However, most of the current algorithms are static monitoring algorithms, which make it difficult to fully extract and process the time-varying dynamic characteristics of the sewage treatment process. In addition, the sewage treatment process has multi-stage characteristics, and the biochemical reactions and dominant parameters in different stages are different, resulting in significant strong dynamic characteristics in the sewage treatment process. The methods for dynamic feature extraction based on time series data mainly include methods based on neural networks, etc.

[0004] However, existing studies have not fully considered the stage characteristics and dynamic characteristics of sewage, resulting in poor diagnostic results for mild sludge bulking failures. Summary of the invention

[0005] The present invention provides a sewage fault diagnosis method and device based on adaptive dynamic slow feature analysis, which is used to solve the defect that the stage characteristics and dynamic characteristics of sewage are not fully considered in the prior art, resulting in poor diagnosis effect of mild sludge bulking faults, and realizes timely and accurate fault judgment.

[0006] The present invention provides a sewage fault diagnosis method based on adaptive dynamic slow feature analysis, comprising: classifying sewage treatment data obtained online based on a neighbor propagation clustering algorithm to obtain multiple clusters of online classified data; inputting the multiple clusters of online classified data into a previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; determining corresponding online superior statistics and online inferior statistics respectively according to the online superior slow features and the online inferior slow features; comparing the online superior statistics with the superior adaptive control limits, and comparing the online inferior statistics with the inferior adaptive control limits, and determining that a fault exists based on the corresponding statistics exceeding the corresponding adaptive control limits, otherwise, determining that no fault exists and the sewage treatment process is operating normally; wherein the adaptive control limits are obtained based on the corresponding online slow features, the corresponding offline slow features and the fixed control limits determined based on the corresponding offline slow features.

[0007] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, multiple clusters of online classified data are input into a previously constructed slow feature analysis model to obtain online dominant slow features and online inferior slow features output by the slow feature analysis model, including: inputting multiple clusters of online classified data into the previously constructed slow feature analysis model to perform data amplification using time window delay for the classified data in each cluster respectively to obtain corresponding amplified data; whitening the amplified data to obtain whitened data; performing singular value decomposition on the whitened data to obtain a corresponding transformation matrix, and determining the corresponding slow features according to the transformation matrix; dividing the slow features based on a preset speed threshold, taking the slow features not greater than the preset speed threshold as the online dominant slow features, and taking the slow features greater than the preset speed threshold as the inferior dominant slow features.

[0008] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, the sewage treatment data acquired online include input variables collected during sewage treatment, and the input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity; before the slow features are divided based on a preset speed threshold, it includes: screening the slow features to eliminate slow features whose speed is faster than the slow features corresponding to the input variables.

[0009] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, before inputting multiple clusters of online classification data into a previously constructed slow feature analysis model, the method includes: obtaining sewage treatment offline data; classifying the sewage treatment offline data based on a neighbor propagation clustering algorithm to obtain multiple clusters of offline classification data; constructing a slow feature analysis model based on the multiple clusters of offline classification data, and using the slow feature analysis model to extract corresponding offline superior slow features and offline inferior slow features; determining corresponding offline superior statistics and offline inferior statistics respectively according to the offline superior slow features and the offline inferior slow features, and determining corresponding superior fixed control limits and inferior fixed control limits in combination with the chi-square distribution.

[0010] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, before comparing the size of the online advantage statistic with the advantage adaptive control limit, and comparing the size of the online disadvantage statistic with the disadvantage adaptive control limit, the method includes: determining the mean of the offline advantage slow feature based on the offline advantage slow feature; determining the advantage difference norm based on the online advantage slow feature and the mean of the offline advantage slow feature; determining the advantage fluctuation value of the online advantage slow feature relative to the offline advantage slow feature according to the advantage difference norm, the online advantage slow feature and the mean of the offline advantage slow feature; obtaining the advantage adaptive control limit according to the advantage fluctuation value and the advantage fixed control limit; determining the mean of the offline disadvantage slow feature based on the offline disadvantage slow feature; determining the disadvantage difference norm based on the online disadvantage slow feature and the mean of the offline disadvantage slow feature; determining the disadvantage fluctuation value of the online disadvantage slow feature relative to the offline disadvantage slow feature according to the disadvantage difference norm, the online disadvantage slow feature and the mean of the offline disadvantage slow feature; obtaining the disadvantage adaptive control limit according to the disadvantage fluctuation value and the disadvantage fixed control limit.

[0011] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, sewage treatment offline data is obtained, including: collecting sewage treatment process data, the sewage treatment process data includes input variables and output variables, the input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity, and the output variables include the concentration of suspended solids in water, biodegradable organic matter, granular inert organic carbon, active heterotrophic bacteria, active autotrophic bacteria, degradable particulate matter, soluble biodegradable organic nitrogen, and total suspended solids; collecting sludge bulking fault data, and adjusting the heterotrophic bacteria attenuation μH value and heterotrophic bacteria growth bH value of benchmark simulation model No. 1 BSM1 to produce fault sample data; according to the sewage treatment process data and the fault sample data, the sewage treatment offline data is obtained, and the sewage treatment offline data is standardized.

[0012] According to a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, based on the nearest neighbor propagation clustering algorithm, the sewage treatment data obtained online are classified to obtain multi-cluster online classification data, including: according to the sewage treatment data obtained online, using the negative Euclidean distance, obtaining the similarity of any two sample points in the sewage treatment data obtained online; initializing the responsibility matrix and the availability matrix between any two sample points; wherein the responsibility matrix is ​​used to characterize the suitability of considering the second sample point as the cluster center for any first sample point, and the availability matrix is ​​used to characterize the possibility of the first sample point considering the second sample point as the cluster center; for any first sample point, according to the availability matrix of the first sample point and other sample points, the similarity of the first sample point and other sample points, and the similarity of the first sample point and the second sample point, the responsibility matrix is ​​updated, and the availability matrix is ​​updated based on the responsibility matrix of other sample points and the second sample point; using a preset damping coefficient, the responsibility matrix and the availability matrix are updated alternately and iteratively, and based on meeting the convergence condition, the update is terminated to obtain the cluster center, and according to the cluster center, the sample points in the sewage treatment data obtained online are allocated to the corresponding clusters to obtain multi-cluster online classification data.

[0013] The present invention also provides a sewage fault diagnosis device based on adaptive dynamic slow feature analysis, comprising: a clustering module, which classifies sewage treatment data obtained online based on a neighbor propagation clustering algorithm to obtain multiple clusters of online classified data; a feature extraction module, which inputs the multiple clusters of online classified data into a previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; a statistical module, which determines the corresponding online superior statistics and online inferior statistics respectively according to the online superior slow features and the online inferior slow features; a fault judgment module, which compares the size of the online superior statistics with the size of the superior adaptive control limit, and compares the size of the online inferior statistics with the size of the inferior adaptive control limit, and determines the existence of a fault based on the corresponding statistics exceeding the corresponding adaptive control limit, otherwise, determines that there is no fault and the operation state of the sewage treatment process is normal; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the sewage fault diagnosis method based on adaptive dynamic slow feature analysis as described in any one of the above methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described sewage fault diagnosis methods based on adaptive dynamic slow feature analysis.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned sewage fault diagnosis methods based on adaptive dynamic slow feature analysis.

[0017] The sewage fault diagnosis method and device based on adaptive dynamic slow feature analysis provided by the present invention classify the sewage treatment data obtained online through the nearest propagation clustering algorithm, so as to automatically determine the appropriate number of clusters based on the similarity between the data without specifying the number of clusters, thereby improving the flexibility in processing complex data and adapting to different data distributions and structures; using the slow feature analysis model, the classified data is subjected to slow feature extraction, and the slow features are divided into dominant slow features that are relatively stable during normal operation and inferior slow features that are prone to appear when disturbed or faulty, so as to fully extract the stage and dynamic characteristics contained in the sewage treatment data; on the basis of long-term dynamic monitoring of the sewage treatment process based on the slow features, the deviation degrees of the dominant slow features and the inferior slow features are quantified by statistics, respectively, so as to more comprehensively reflect the changes in the state of the sewage treatment process; and by comparing the statistics with the size of the adaptive control limit, it is timely judged whether there is a fault in the sewage treatment process, thereby improving the accuracy of fault determination, improving the sensitivity of fault diagnosis, and effectively monitoring the sewage treatment process in real time, preventing the sewage treatment effluent water quality from exceeding the standard due to faults such as sludge bulking, and reducing the losses caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention; Figure 2 It is a graph of the variation of the statistical quantity of the advantage slow characteristic of the online test process provided by the present invention; Figure 3 It is a graph of the variation of the statistics of the slow characteristic of the online test process provided by the present invention; Figure 4It is a structural schematic diagram of a sewage fault diagnosis device based on adaptive dynamic slow feature analysis provided by the present invention; Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Figure 1 is a flow chart of a sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the present invention, such as Figure 1 As shown, the method includes: S11, based on the neighbor propagation clustering algorithm, the sewage treatment data obtained online is classified to obtain multiple clusters of online classification data; S12, inputting the multi-cluster online classification data into the previously constructed slow feature analysis model to obtain the online superior slow features and the online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; S13, determining corresponding online advantage statistics and online disadvantage statistics respectively according to the online advantage slow feature and the online disadvantage slow feature; S14, comparing the online advantage statistic with the advantage adaptive control limit, and comparing the online disadvantage statistic with the disadvantage adaptive control limit, and determining that a fault exists based on the corresponding statistic exceeding the corresponding adaptive control limit, otherwise, determining that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0022] It should be noted that the step numbers "S1N" in this specification do not represent the order of the sewage fault diagnosis method based on adaptive dynamic slow feature analysis. Figure 2-3 The sewage fault diagnosis method based on adaptive dynamic slow feature analysis of the present invention is described.

[0023] Step S11, based on the nearest neighbor propagation clustering algorithm, classify the sewage treatment data obtained online to obtain multiple clusters of online classified data.

[0024] In this embodiment, based on the neighbor propagation clustering algorithm, the sewage treatment data obtained online is classified to obtain multiple clusters of online classified data, including: S111, according to the sewage treatment data obtained online, using negative Euclidean distance, obtain the similarity between any two sample points in the sewage treatment data obtained online.

[0025] It should be added that any two sample points in the sewage treatment data obtained online and Similarity , expressed as:

[0026] S112, initialize the responsibility matrix and availability matrix between any two sample points; wherein the responsibility matrix is ​​used to characterize the suitability of the first sample point for considering the second sample point as the cluster center, and the availability matrix is ​​used to characterize the possibility of the first sample point considering the second sample point as the cluster center.

[0027] In this embodiment, the sample points and The responsibility matrix between is represented as r(i,k), and the sample points and The availability matrix between them is represented as a(i,k).

[0028] S113, for any first sample point, update the responsibility matrix according to the availability matrix of the first sample point and other sample points, the similarity between the first sample point and other sample points, and the similarity between the first sample point and the second sample point, and update the availability matrix based on the responsibility matrix of any other sample point and the second sample point.

[0029] It should be added that the updated responsibility matrix is ​​expressed as:

[0030] in, Represents the first sample point With the second sample point The similarity of Represents the first sample point and other sample points The availability matrix between Represents the first sample point and other sample points The similarity between .

[0031] The updated availability matrix is ​​expressed as:

[0032]

[0033] in, Represents the second sample point The responsibility matrix can be based on other sample points With the second sample point The maximum value between the responsibility matrix and 0 is determined; Indicates other sample points With the second sample point Responsibility matrix.

[0034] S114, using the preset damping coefficient, alternately iteratively update the responsibility matrix and the availability matrix, and end the update based on meeting the convergence condition to obtain the cluster center, and according to the cluster center, assign the sample points in the sewage treatment data obtained online to the corresponding cluster to obtain multi-cluster online classification data.

[0035] In this embodiment, the preset damping coefficient is used to alternately iterate and update the responsibility matrix and the availability matrix, which are respectively expressed as:

[0036]

[0037] in, ' represents the first sample point updated after the mth iteration and the second sample point Responsibility matrix between Indicates the preset damping coefficient, which can be set based on prior experience or actual design requirements; Represents the first sample point obtained after the mth iteration and the second sample point Responsibility matrix between Represents the first sample point updated after the mth iteration and the second sample point Availability matrix between ; Represents the first sample point obtained after the mth iteration and the second sample point The availability matrix between .

[0038] It should be noted that the convergence condition can be set according to the actual design requirements, such as matrix convergence or reaching a predetermined number of iterations. Through the neighbor propagation AP clustering algorithm, the cluster center can be determined by the similarity between data points and the transmitted information to divide the collected data into different clusters, and obtain multi-cluster online classification data, which is expressed as: ,in, represents the i-th sewage treatment stage, and B represents the number of stages of sewage treatment data stratification, that is, the number of clusters.

[0039] Step S12, inputting the multi-cluster online classification data into the previously constructed slow feature analysis model to obtain the online superior slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features.

[0040] In this embodiment, multiple clusters of online classification data are input into the previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model, including: S121, inputting the multi-cluster online classification data into the previously constructed slow feature analysis model, so as to amplify the classification data in each cluster respectively by using the time window delay to obtain corresponding amplified data.

[0041] It should be noted that for the data in each cluster, in order to better describe its process dynamic characteristics, the time window delay method can be used to amplify all the data at the current moment and within the next d moments to obtain the amplified data , expressed as:

[0042] S122, performing whitening processing on the amplified data to obtain whitened data.

[0043] It should be noted that the whitening process of the amplification data includes: , and obtain the outer product of the amplification matrix ; The outer product of the amplification matrix Perform whitening, that is, calculate , where U is The eigenvector matrix of is a diagonal matrix with diagonal elements The singular values ​​of . After whitening by whitening matrix Q, the whitened data is obtained ,in, , represents the J×J identity matrix.

[0044] S123, performing singular value decomposition on the whitened data to obtain a corresponding transformation matrix, and determining a corresponding slow feature according to the transformation matrix.

[0045] It should be noted that the singular value decomposition of the whitened data is performed to obtain the corresponding transformation matrix, including: performing singular value decomposition on the covariance matrix of the first-order difference of the whitened data, that is, , where P is The eigenvector matrix of , yes According to the singular value decomposition results, the transformation matrix is ​​obtained , where X represents the amplification matrix, W e Represents the inferior transformation matrix; According to the transformation matrix, the slow feature is obtained .

[0046] S124, based on the preset speed threshold, the slow features are divided, and the slow features that are not greater than the preset speed threshold are used as online dominant slow features. The dominant slow feature conversion matrix is ​​W d , the slow features greater than the preset speed threshold are regarded as inferior superior slow features, and the inferior slow feature conversion matrix is ​​W e .

[0047] It should be noted that the dominant slow feature is expressed as , the disadvantageous slow feature is expressed as In addition, the dominant slow features represent relatively stable and important features during the normal operation of sewage treatment, while the inferior slow features are more susceptible to interference or are the first to become abnormal when a fault occurs. By dividing the slow features into dominant slow features and inferior slow features, a more targeted analysis of different aspects of sewage treatment can be carried out, which helps to more accurately locate the source and nature of the fault in the future.

[0048] In addition, after obtaining the online superior slow feature and the online inferior slow feature, the method further includes: using the conversion matrix corresponding to the online superior slow feature as the online superior conversion matrix, and using the conversion matrix corresponding to the online inferior slow feature as the online inferior conversion matrix; and determining the number of the online superior slow feature and the online inferior slow feature, which is specifically expressed as:

[0049]

[0050] in, represents the number of inferior slow features; M represents the number of superior slow features; Indicates the speed of transformation of each column of slow feature data; Indicates the speed of change of each column of the original amplified data; m indicates the dimension of the original data; d indicates the dimension of the sample to be amplified, Represents the total dimension of the sample after augmentation.

[0051] In an optional embodiment, the sewage treatment data obtained online includes input variables collected by sewage treatment, and the input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity. Accordingly, before the slow features are divided based on a preset speed threshold, it includes: screening the slow features to eliminate the slow features whose speed is faster than the slow features corresponding to the input variables.

[0052] In an optional embodiment, before inputting multiple clusters of online classification data into a previously constructed slow feature analysis model, the method includes: obtaining sewage treatment offline data; classifying the sewage treatment offline data based on a neighbor propagation clustering algorithm to obtain multiple clusters of offline classification data; constructing a slow feature analysis model based on the multiple clusters of offline classification data, and using the slow feature analysis model to extract corresponding offline dominant slow features and offline inferior slow features; determining corresponding offline dominant statistics and offline inferior statistics respectively according to the offline dominant slow features and the offline inferior slow features, and determining corresponding dominant fixed control limits and inferior fixed control limits in combination with the chi-square distribution.

[0053] It should be noted that based on the neighbor propagation clustering algorithm, the classification of sewage treatment offline data can refer to step S11; using the slow feature analysis model to extract the corresponding offline advantage slow features and offline disadvantage slow features, refer to step S12; no repetition will be made here.

[0054] Specifically, obtaining offline data of sewage treatment includes: collecting sewage treatment process data, sewage treatment process data includes input variables and output variables, input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity, output variables include the concentration of suspended solids in water, biodegradable organic matter, granular inert organic carbon, active heterotrophic bacteria, active autotrophic bacteria, degradable particulate matter, soluble biodegradable organic nitrogen, total suspended solids; collecting sludge bulking fault data, and producing fault sample data by adjusting the heterotrophic bacteria attenuation μH value and heterotrophic bacteria growth bH value of benchmark simulation model No. 1 BSM1; obtaining sewage treatment offline data based on sewage treatment process data and fault sample data, and standardizing the sewage treatment offline data.

[0055] It should be noted that the data of input variables and output variables can be obtained based on the BSM1 model. In addition, the standardized data is expressed as ,in, , represents the mean of the original data, Represents the variance of the original data.

[0056] In addition, according to the offline advantage slow feature and the offline disadvantage slow feature, the corresponding offline advantage statistics and offline disadvantage statistics are determined respectively, and combined with the chi-square distribution, the corresponding advantage fixed control limits and disadvantage fixed control limits are determined, including: for the advantage slow feature and the disadvantage slow feature, Hotelling T 2 Statistics, get offline advantage statistics and offline disadvantage statistics; based on preset components and offline advantage statistics and offline disadvantage statistics, get advantage fixed control limits and disadvantage fixed control limits.

[0057] It should be added that the offline advantage statistic is expressed as , and the offline disadvantage statistics are expressed as , What is measured is the static change of the slowest main slow feature across the subspace, which is taken as the statistic of the dominant slow feature; The measured quantity is the static change of the other fastest and slowest features, which is used as the statistic of the inferior slow feature. , assuming is an independent Gaussian distribution, then The statistic is subject to the degrees of freedom of distributed, The statistic is subject to the degrees of freedom of Distribution, that is:

[0058]

[0059] In addition, the advantage is to fix the control limits, expressed as:

[0060]

[0061]

[0062]

[0063]

[0064] in, represents the advantage fixed control limit; represents the quantile corresponding to the (1-α) confidence level in the chi-square distribution, where α represents the confidence level; express and The scaling factor combines information from the dominant slow feature and the sample-wide Hotelling's T² statistic to determine a scaling factor in the dominant fixed control limits; express and The proportionality coefficient, when determining the fixed control limits of Hotelling's T² statistic, is and Work together to determine the specific values ​​of the control limits; It represents the overall average level of the sample in terms of Hotelling's T² statistic, which is an important reference value for constructing fixed control limits; represents the advantage statistic; n represents the sample size; Indicates the variance of the sample; generally .

[0065] The inferior fixed control limits are expressed as:

[0066]

[0067]

[0068]

[0069]

[0070] in, represents the advantage fixed control limit; represents the quantile corresponding to the (1-α) confidence level in the chi-square distribution, where α represents the confidence level; express and The scaling factor combines information from the dominant slow feature and the sample-wide Hotelling's T² statistic to determine a scaling factor in the dominant fixed control limits; express and The proportionality coefficient, when determining the fixed control limits of Hotelling's T² statistic, is and Work together to determine the specific values ​​of the control limits; It represents the overall average level of the sample in terms of Hotelling's T² statistic, which is an important reference value for constructing fixed control limits; represents the advantage statistic; n represents the sample size; Indicates the variance of the sample. In general, .

[0071] Step S13, according to the online advantage slow feature and the online disadvantage slow feature, respectively determine the corresponding online advantage statistics and online disadvantage statistics. It should be noted that the online advantage statistics and online disadvantage statistics can refer to the above-mentioned offline advantage statistics and offline disadvantage statistics, which will not be repeated here. In addition, through Hotelling T 2 Statistics quantify the degree of deviation of each dominant slow feature and inferior slow feature to consider the interrelationship between multiple features, so as to more comprehensively reflect the changes in sewage treatment status and more sensitively detect abnormal conditions in the sewage treatment process.

[0072] Step S14, compare the online advantage statistic with the advantage adaptive control limit, and compare the online disadvantage statistic with the disadvantage adaptive control limit, and determine that a fault exists based on the corresponding statistic exceeding the corresponding adaptive control limit, otherwise, determine that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0073] It should be noted that by comparing the size of the statistic with the adaptive control limit, a clear judgment standard is provided to reduce the uncertainty of subjective judgment, so as to accurately judge whether there is a fault in the system. If the statistic exceeds the control limit, it means that the current state of the sewage treatment process has deviated from the normal range, and the possibility of a fault is high, which improves the accuracy of fault determination.

[0074] In an optional embodiment, before comparing the online advantage statistic with the size of the advantage adaptive control limit, and comparing the online disadvantage statistic with the size of the disadvantage adaptive control limit, it includes: determining the mean of the offline advantage slow feature based on the offline advantage slow feature; determining the advantage difference norm based on the online advantage slow feature and the mean of the offline advantage slow feature; determining the advantage fluctuation value of the online advantage slow feature relative to the offline advantage slow feature according to the advantage difference norm, the online advantage slow feature and the mean of the offline advantage slow feature; and obtaining the advantage adaptive control limit according to the advantage fluctuation value and the advantage fixed control limit.

[0075] Similarly, before comparing the online advantage statistic with the advantage adaptive control limit, and comparing the online disadvantage statistic with the disadvantage adaptive control limit, it also includes: determining the mean of the offline disadvantage slow feature based on the offline disadvantage slow feature; determining the disadvantage difference norm based on the online disadvantage slow feature and the mean of the offline disadvantage slow feature; determining the disadvantage fluctuation value of the online disadvantage slow feature relative to the offline disadvantage slow feature according to the disadvantage difference norm, the online disadvantage slow feature and the mean of the offline disadvantage slow feature; and obtaining the disadvantage adaptive control limit according to the disadvantage fluctuation value and the disadvantage fixed control limit.

[0076] It is worth noting that by considering the deviation between the online slow feature and the average value of the offline slow feature as a weight and combining it with the fixed control limit calculated in the offline process, the control limit is adaptively adjusted to better reflect the difference between the current state and the normal state of the sewage treatment process, so that fault detection can better adapt to different working conditions and environments.

[0077] It should be added that the difference norm is expressed as:

[0078]

[0079] in, Indicates the corresponding online slow feature; Indicates the corresponding offline slow feature mean; Represents the corresponding offline slow feature; t represents the sampling time, t∈1,…,m.

[0080] In addition, at the tth sampling, the fluctuation value of the online slow feature relative to the corresponding offline slow feature , expressed as:

[0081] When a sudden failure occurs, ,but , Indicates variables, so Similarly, when the online data are all normal and are all smaller than the mean of the normal data, Therefore, the adaptive coefficient to set the adaptive control limits.

[0082] It should be noted that if the final calculation is the dominant adaptive control limit, the slow feature in the above formula is the corresponding dominant slow feature, otherwise it is the corresponding inferior slow feature.

[0083] In addition, the advantage of adaptive control limits and inferior adaptive control limits , respectively expressed as:

[0084]

[0085] In an optional embodiment, the method further includes: evaluating the performance of fault detection using indicators such as accuracy, fault detection rate, and false alarm rate, where the accuracy, fault detection rate, and false alarm rate indicators are respectively expressed as:

[0086]

[0087]

[0088] Among them, TP (True Positives) represents true positives, that is, the number of positive examples correctly judged as positive and negative examples; FP (False Positives) represents false positives, that is, the number of negative examples mistakenly judged as positive examples; FN (False Negatives) represents false negatives, that is, the number of positive examples mistakenly judged as negative examples; TN (True Negatives) represents true negatives, that is, the number of negative examples correctly judged as negative examples.

[0089] In an optional embodiment, in order to better verify the effect of the present invention in industrial process fault detection, a data set obtained from the BSM1 simulation platform developed by the International Water Association was selected to verify the effect of the present invention.

[0090] Sludge bulking is a common type of failure in the sewage treatment process. When sludge bulking occurs, the sedimentation and compression characteristics of the sludge deteriorate, the suspended solids increase, the sludge loss is serious, and the biochemical system may fail, which directly threatens the normal operation of the sewage treatment system. Toxic shock failure is caused by the discharge of a large amount of toxic sewage, which leads to a decrease in the microbial activity of the activated sludge and even the failure of the sewage biochemical reaction system. Retention failure occurs when the normal growth of heterotrophic organisms decreases. This experiment can produce different degrees of retention failure by adjusting the parameters of heterotrophic bacterial growth rate μH and heterotrophic attenuation coefficient bH. Table 1 describes the 6 types of sludge bulking failures involved in this research experiment. Table 2 provides the sewage process process parameters of some sewage treatment process variables.

[0091] During the implementation process, 14 days of offline training data were collected based on the BSM1 simulation platform, with a total of 1,344 samples. The online data used was data from BSM1 running in normal conditions for 2 days under three different working conditions (DRY, RAIN, and STORM), and then running continuously for 7 days under fault conditions. A total of 864 samples were collected for each fault type, and the fault occurred between time 192 and time 864.

[0092] Table 1. Types of fault data generated by BSM1

[0093] Table 2 Variable names collected by BSM1

[0094] The stage division results based on the AP clustering algorithm under three different working conditions are shown in Table 3. The stage division results basically correspond to the two anoxic pools and three aerobic pools in the biochemical reaction pool, which also illustrates the effectiveness of the stage division results.

[0095] Table 3 Results of stage division of sewage treatment plants

[0096] The statistics of the good and bad slow features are compared with the adaptive control limits. When the statistics exceed the corresponding adaptive control limits, it indicates that a fault has occurred in the process; otherwise, the process is considered to be in a normal state.

[0097] Figure 2 and Figure 3This is the monitoring result of sludge bulking failure of sewage treatment plant based on multi-level adaptive dynamic slow feature analysis model proposed by the present invention, in which Figure 2 is the fault detection result in the dominant slow feature subspace, Figure 3 This is the fault detection result of the inferior slow feature. Figure 2 and Figure 3 According to the experimental results, during the online test phase, the sewage treatment plant operated normally in the first two days, but then the statistics exceeded the adaptive threshold, indicating that a failure occurred in the sewage treatment process. As can be seen from the figure, the model based on multi-level adaptive dynamic slow feature analysis developed by the present invention has superior performance.

[0098] In summary, the embodiment of the present invention classifies the sewage treatment data obtained online through the neighbor propagation clustering algorithm, so as to automatically determine the appropriate number of clusters based on the similarity between the data without specifying the number of clusters, thereby improving the flexibility in processing complex data to adapt to different data distributions and structures; using the slow feature analysis model, the classified data is subjected to slow feature extraction, and the slow features are divided into dominant slow features that are relatively stable during normal operation and inferior slow features that are prone to appear when disturbed or faulty, so as to fully extract the stage and dynamic characteristics contained in the sewage treatment data, and on the basis of long-term dynamic monitoring of the sewage treatment process based on slow features, the deviation degree of the dominant slow features and the inferior slow features is quantified by statistics, respectively, to more comprehensively reflect the changes in the state of the sewage treatment process, and by comparing the statistics with the size of the adaptive control limit, it is timely judged whether there is a fault in the sewage treatment process, thereby improving the accuracy of fault determination, improving the sensitivity of fault diagnosis, and effectively monitoring the sewage treatment process in real time, preventing the sewage treatment effluent water quality from exceeding the standard due to faults such as sludge bulking, and reducing the losses caused by faults.

[0099] The sewage fault diagnosis device based on adaptive dynamic slow feature analysis provided by the present invention is described below. The sewage fault diagnosis device based on adaptive dynamic slow feature analysis described below and the sewage fault diagnosis method based on adaptive dynamic slow feature analysis described above can be referenced to each other.

[0100] Figure 4 The schematic diagram of the structure of a sewage fault diagnosis device based on adaptive dynamic slow feature analysis is shown, and the device comprises: The clustering module 41 classifies the sewage treatment data obtained online based on the nearest neighbor propagation clustering algorithm to obtain multiple clusters of online classified data; The feature extraction module 42 inputs the multi-cluster online classification data into the previously constructed slow feature analysis model to obtain the online superior slow features and the online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; The statistical module 43 determines the corresponding online advantage statistics and online disadvantage statistics according to the online advantage slow feature and the online disadvantage slow feature; The fault judgment module 44 compares the size of the online advantage statistic with the advantage adaptive control limit, and compares the size of the online disadvantage statistic with the disadvantage adaptive control limit, and determines that a fault exists based on the corresponding statistic exceeding the corresponding adaptive control limit, otherwise, it is determined that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0101] In this embodiment, the clustering module 41 includes: a similarity determination unit, which uses negative Euclidean distance to obtain the similarity between any two sample points in the sewage treatment data obtained online; an initialization unit, which initializes the responsibility matrix and the availability matrix between any two sample points; wherein the responsibility matrix is ​​used to characterize the suitability of the second sample point as the cluster center for the first sample point, and the availability matrix is ​​used to characterize the possibility of the first sample point considering the second sample point as the cluster center; a matrix update unit, which updates the responsibility matrix for any first sample point according to the availability matrix of the first sample point and other sample points, the similarity of the first sample point and other sample points, and the similarity of the first sample point and the second sample point, and updates the availability matrix based on the responsibility matrix of any other sample point and the second sample point; a clustering unit, which uses a preset damping coefficient to alternately iteratively update the responsibility matrix and the availability matrix, and terminates the update based on meeting the convergence condition to obtain the cluster center, and according to the cluster center, allocates the sample points in the sewage treatment data obtained online to the corresponding clusters to obtain multi-cluster online classification data.

[0102] The feature extraction module 42 includes: a data amplification unit, which inputs multiple clusters of online classification data into a previously constructed slow feature analysis model, so as to perform data amplification using a time window delay for the classification data in each cluster, and obtain corresponding amplified data; a whitening processing unit, which performs whitening processing on the amplified data, and obtains whitened data; a feature extraction unit, which performs singular value decomposition on the whitened data, obtains a corresponding transformation matrix, and determines the corresponding slow features based on the transformation matrix; a feature division unit, which divides the slow features based on a preset speed threshold, and uses the slow features not greater than the preset speed threshold as the online dominant slow features, and uses the slow features greater than the preset speed threshold as the inferior dominant slow features.

[0103] In an optional embodiment, the feature extraction module 42 further includes: a matrix division unit, after obtaining the online superior slow feature and the online inferior slow feature, using the conversion matrix corresponding to the online superior slow feature as the online superior conversion matrix, and using the conversion matrix corresponding to the online inferior slow feature as the online inferior conversion matrix; a quantity determination unit, determining the quantity of the online superior slow feature and the online inferior slow feature.

[0104] In an optional embodiment, the feature extraction module 42 further includes: a feature screening unit, which screens the slow features before dividing the slow features based on a preset speed threshold, and eliminates the slow features that are faster than the slow features corresponding to the input variables.

[0105] In an optional embodiment, the device also includes: an offline data acquisition module, which obtains sewage treatment offline data before inputting the multi-cluster online classification data into the previously constructed slow feature analysis model; a clustering module, which classifies the sewage treatment offline data based on the nearest neighbor propagation clustering algorithm to obtain multi-cluster offline classification data; a feature extraction module, which constructs a slow feature analysis model based on the multi-cluster offline classification data, and uses the slow feature analysis model to extract corresponding offline dominant slow features and offline inferior slow features; a statistical module, which determines the corresponding offline dominant statistics and offline inferior statistics according to the offline dominant slow features and the offline inferior slow features, and determines the corresponding dominant fixed control limits and inferior fixed control limits in combination with the chi-square distribution.

[0106] Specifically, the offline data acquisition module includes: a first data acquisition unit, which collects sewage treatment process data, the sewage treatment process data includes input variables and output variables, the input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity, and the output variables include the concentration of suspended solids in water, biodegradable organic matter, granular inert organic carbon, active heterotrophic bacteria, active autotrophic bacteria, degradable particulate matter, soluble biodegradable organic nitrogen, and total suspended solids. At least one; a second data acquisition unit, which collects sludge bulking fault data, and adjusts the heterotrophic bacteria attenuation μH value and heterotrophic bacteria growth bH value of the benchmark simulation model No. 1 BSM1 to produce fault sample data; a standardization processing unit, which obtains sewage treatment offline data according to the sewage treatment process data and the fault sample data, and performs standardization processing on the sewage treatment offline data.

[0107] In addition, the statistical module includes: a statistical unit, which constructs Hotelling T for the superior slow feature and the inferior slow feature respectively. 2 Statistics, to obtain offline advantage statistics and offline disadvantage statistics; a control limit determination unit, based on preset components and the offline advantage statistics and offline disadvantage statistics, to obtain advantage fixed control limits and disadvantage fixed control limits.

[0108] In an optional embodiment, the device further includes: a mean determination module, which determines the mean of the offline dominant slow feature based on the offline dominant slow feature before comparing the online dominant statistic with the size of the dominant adaptive control limit and comparing the online inferior statistic with the size of the inferior adaptive control limit; a norm determination module, which determines the dominant difference norm based on the online dominant slow feature and the mean of the offline dominant slow feature; a fluctuation determination module, which determines the dominant fluctuation value of the online dominant slow feature relative to the offline dominant slow feature according to the dominant difference norm, the online dominant slow feature and the mean of the offline dominant slow feature; and a control limit adjustment module, which obtains the dominant adaptive control limit according to the dominant fluctuation value and the dominant fixed control limit.

[0109] Similarly, the device also includes: a mean determination module, which determines the mean of the offline inferior slow feature based on the offline inferior slow feature before comparing the online superiority statistic with the superiority adaptive control limit and comparing the online inferiority statistic with the inferiority adaptive control limit; a norm determination module, which determines the inferior difference norm based on the online inferior slow feature and the mean of the offline inferior slow feature; a fluctuation determination module, which determines the inferior fluctuation value of the online inferior slow feature relative to the offline inferior slow feature according to the inferior difference norm, the online inferior slow feature and the mean of the offline inferior slow feature; and a control limit adjustment module, which obtains the inferior adaptive control limit according to the inferior fluctuation value and the inferior fixed control limit.

[0110] In an optional embodiment, the device further includes: an evaluation module for evaluating the performance of fault detection using indicators such as accuracy, fault detection rate and false alarm rate.

[0111] In summary, the embodiment of the present invention uses the nearest neighbor propagation clustering algorithm through the clustering module to classify the sewage treatment data obtained online, so as to automatically determine the appropriate number of clusters based on the similarity between the data without specifying the number of clusters, thereby improving the flexibility in processing complex data to adapt to different data distributions and structures; the feature extraction module uses the slow feature analysis model to extract slow features from the classified data, and divides the slow features into dominant slow features that are relatively stable during normal operation and inferior slow features that are prone to appear when disturbed or faulty, so as to fully extract the stage and dynamic characteristics contained in the sewage treatment data, and on the basis of long-term dynamic monitoring of the sewage treatment process based on slow features, the deviation degree of the dominant slow features and the inferior slow features is quantified by the statistical module statistics, respectively, so as to more comprehensively reflect the change of the state of the sewage treatment process, and the fault judgment module compares the statistics with the size of the adaptive control limit, and timely judges whether there is a fault in the sewage treatment process, thereby improving the accuracy of fault determination, improving the sensitivity of fault diagnosis, and effectively monitoring the sewage treatment process in real time, preventing the sewage treatment effluent water quality from exceeding the standard due to faults such as sludge swelling, and reducing the losses caused by faults.

[0112] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 530 to execute the sewage fault diagnosis method based on adaptive dynamic slow feature analysis, which includes: based on the nearest neighbor propagation clustering algorithm, classifying the sewage treatment data obtained online to obtain multiple clusters of online classification data; inputting the multiple clusters of online classification data into the slow feature analysis model constructed in advance to obtain the online dominant slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline dominant slow features and offline inferior slow features; according to the online dominant slow features and the online inferior slow features, respectively determine the corresponding online dominant statistics and online inferior statistics; compare the size of the online dominant statistics with the size of the dominant adaptive control limit, and compare the size of the online inferior statistics with the size of the inferior adaptive control limit, and determine the existence of a fault based on the corresponding statistics exceeding the corresponding adaptive control limit, otherwise, determine that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0113] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0114] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the above methods, the method comprising: classifying the sewage treatment data acquired online based on the nearest neighbor propagation clustering algorithm to obtain multiple clusters of online classified data; inputting the multiple clusters of online classified data into a previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model; wherein, the slow feature analysis The model is constructed based on offline data of sewage treatment and its corresponding offline superior slow characteristics and offline inferior slow characteristics; according to the online superior slow characteristics and the online inferior slow characteristics, the corresponding online superior statistics and online inferior statistics are determined respectively; the online superior statistics are compared with the superior adaptive control limits, as well as the online inferior statistics and the inferior adaptive control limits, and based on the corresponding statistics exceeding the corresponding adaptive control limits, it is determined that there is a fault, otherwise, it is determined that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limits are obtained based on the corresponding online slow characteristics, the corresponding offline slow characteristics and the fixed control limits determined based on the corresponding offline slow characteristics.

[0115] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the sewage fault diagnosis method based on adaptive dynamic slow feature analysis provided by the above methods, the method comprising: classifying the sewage treatment data acquired online based on the nearest propagation clustering algorithm to obtain multiple clusters of online classified data; inputting the multiple clusters of online classified data into a previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is based on the sewage treatment offline data and its The corresponding offline superior slow feature and offline inferior slow feature are constructed; according to the online superior slow feature and the online inferior slow feature, the corresponding online superior statistic and the online inferior statistic are determined respectively; the online superior statistic and the superior adaptive control limit are compared, as well as the online inferior statistic and the inferior adaptive control limit are compared, and based on the corresponding statistic exceeding the corresponding adaptive control limit, it is determined that there is a fault, otherwise, it is determined that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

[0116] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sewage fault diagnosis method based on adaptive dynamic slow feature analysis, characterized in that: include: Based on the neighbor propagation clustering algorithm, the sewage treatment data obtained online are classified to obtain multiple clusters of online classification data; Inputting the multi-cluster online classification data into the previously constructed slow feature analysis model to obtain the online superior slow features and the online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; According to the online advantage slow feature and the online disadvantage slow feature, respectively determine the corresponding online advantage statistics and online disadvantage statistics; Compare the sizes of the online advantage statistic and the advantage adaptive control limit, and compare the sizes of the online disadvantage statistic and the disadvantage adaptive control limit, and determine that a fault exists based on the corresponding statistic exceeding the corresponding adaptive control limit, otherwise, determine that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

2. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 1 is characterized in that: Inputting the multi-cluster online classification data into the previously constructed slow feature analysis model to obtain online superior slow features and online inferior slow features output by the slow feature analysis model includes: Inputting the multi-cluster online classification data into the previously constructed slow feature analysis model, so as to perform data amplification using time window delay for the classification data in each cluster, respectively, to obtain corresponding amplified data; Performing whitening processing on the amplified data to obtain whitened data; Performing singular value decomposition on the whitened data to obtain a corresponding transformation matrix, and determining a corresponding slow feature according to the transformation matrix; Based on a preset speed threshold, the slow features are divided, and the slow features not greater than the preset speed threshold are taken as online dominant slow features, and the slow features greater than the preset speed threshold are taken as inferior dominant slow features.

3. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 2 is characterized in that: The sewage treatment data obtained online includes input variables collected during sewage treatment, and the input variables include dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity; Before the slow feature is divided based on the preset speed threshold, the method includes: The slow features are screened to eliminate slow features that are faster than the slow features corresponding to the input variables.

4. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 1 is characterized in that: Before inputting the multi-cluster online classification data into the previously constructed slow feature analysis model, the process includes: Obtain offline data of sewage treatment; Based on the neighbor propagation clustering algorithm, the sewage treatment offline data is classified to obtain multiple clusters of offline classified data; Based on the multi-cluster offline classification data, a slow feature analysis model is constructed, and the corresponding offline advantage slow features and offline disadvantage slow features are extracted using the slow feature analysis model; According to the offline advantage slow feature and the offline disadvantage slow feature, the corresponding offline advantage statistics and offline disadvantage statistics are determined respectively, and in combination with the chi-square distribution, the corresponding advantage fixed control limits and disadvantage fixed control limits are determined.

5. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 4 is characterized in that: Before comparing the online superiority statistic with the superiority adaptive control limit, and comparing the online inferiority statistic with the inferiority adaptive control limit, the method includes: Based on the offline dominant slow feature, determining the offline dominant slow feature mean; Determining a dominant difference norm based on the online dominant slow feature and the mean of the offline dominant slow feature; Determine, according to the advantage difference norm, the online advantage slow feature and the mean of the offline advantage slow feature, the advantage fluctuation value of the online advantage slow feature relative to the offline advantage slow feature; Obtaining a dominant adaptive control limit according to the dominant fluctuation value and the dominant fixed control limit; Based on the offline disadvantage slow feature, determining the offline disadvantage slow feature mean; Determining a disadvantage difference norm based on the online disadvantage slow feature and the offline disadvantage slow feature mean; Determining a disadvantage fluctuation value of the online disadvantage slow feature relative to the offline disadvantage slow feature according to the disadvantage difference norm, the online disadvantage slow feature and the mean of the offline disadvantage slow feature; According to the inferior fluctuation value and the inferior fixed control limit, the inferior adaptive control limit is obtained.

6. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 4 is characterized in that: The obtaining of offline sewage treatment data includes: Collecting sewage treatment process data, the sewage treatment process data including input variables and output variables, the input variables including dissolved oxygen, nitrate and nitrite, ammonia nitrogen, ion concentration and alkalinity, the output variables including the concentration of suspended solids in water, biodegradable organic matter, particulate inert organic carbon, active heterotrophic bacteria, active autotrophic bacteria, degradable particulate matter, soluble biodegradable organic nitrogen, and total suspended solids; Collect sludge bulking failure data, and produce failure sample data by adjusting the heterotrophic bacteria decay μH value and heterotrophic bacteria growth bH value of the benchmark simulation model No. 1 BSM1; According to the sewage treatment process data and the fault sample data, sewage treatment offline data is obtained, and the sewage treatment offline data is standardized.

7. The sewage fault diagnosis method based on adaptive dynamic slow feature analysis according to claim 1 is characterized in that: The sewage treatment data obtained online is classified based on the nearest neighbor propagation clustering algorithm to obtain multiple clusters of online classified data, including: According to the sewage treatment data obtained online, using negative Euclidean distance to obtain the similarity between any two sample points in the sewage treatment data obtained online; Initialize the responsibility matrix and availability matrix between any two sample points; wherein the responsibility matrix is ​​used to characterize the suitability of considering the second sample point as the cluster center for any first sample point, and the availability matrix is ​​used to characterize the possibility that the first sample point considers the second sample point as the cluster center; For any of the first sample points, the responsibility matrix is ​​updated according to the availability matrix between the first sample point and other sample points, the similarity between the first sample point and the other sample points, and the similarity between the first sample point and the second sample point, and the availability matrix is ​​updated based on the responsibility matrix between the other sample points and the second sample point; Using the preset damping coefficient, the responsibility matrix and the availability matrix are updated alternately and iteratively, and based on meeting the convergence condition, the update is terminated to obtain the cluster center, and according to the cluster center, the sample points in the sewage treatment data obtained online are allocated to the corresponding clusters to obtain multi-cluster online classification data.

8. A sewage fault diagnosis device based on adaptive dynamic slow feature analysis, characterized in that: include: The clustering module classifies the sewage treatment data obtained online based on the nearest neighbor propagation clustering algorithm to obtain multiple clusters of online classified data; A feature extraction module, inputting the multi-cluster online classification data into a previously constructed slow feature analysis model, and obtaining online superior slow features and online inferior slow features output by the slow feature analysis model; wherein the slow feature analysis model is constructed based on the sewage treatment offline data and its corresponding offline superior slow features and offline inferior slow features; A statistical module, which determines corresponding online advantage statistics and online disadvantage statistics according to the online advantage slow feature and the online disadvantage slow feature; A fault judgment module compares the size of the online advantage statistic with the advantage adaptive control limit, and compares the size of the online disadvantage statistic with the disadvantage adaptive control limit, and determines that a fault exists based on the corresponding statistic exceeding the corresponding adaptive control limit, otherwise, it is determined that there is no fault and the sewage treatment process is operating normally; wherein the adaptive control limit is obtained based on the corresponding online slow feature, the corresponding offline slow feature and the fixed control limit determined based on the corresponding offline slow feature.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the sewage fault diagnosis method based on adaptive dynamic slow feature analysis as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the sewage fault diagnosis method based on adaptive dynamic slow feature analysis as described in any one of claims 1 to 7 is implemented.

Images