Method for extracting evolution law of complex system under multi-failure coupling working condition

Abstract

The method for extracting evolution law of complex system under multiple fault coupling conditions comprises the following steps: S1, constructing historical state information dataset and collecting incremental data in real time; S2, preprocessing the dataset and the incremental data to obtain multiple sets of normalized data; S3, performing clustering analysis on the multiple sets of normalized data to obtain clustering results, and determining the number N of categories and the centers of the categories; S4, performing fault backtracking on the data in S3 by using the number N of categories and the centers of the categories to obtain multiple evolution clustering results; S5, extracting law from each clustering result to obtain multiple optimized evolution clustering results; S6, performing fault positioning and prediction according to the optimized evolution clustering results; and S7, performing cause analysis according to the correlation between the prediction results and the measured data in S6. The evolution law extraction method solves the problems of fault prediction and tracing under the condition of coupled faults.

Description

Technical Field

[0001] This invention belongs to the field of mechanical system fault detection and diagnosis technology, specifically involving a method for extracting the evolution law of complex systems under multiple fault coupling conditions. Background Technology

[0002] With the continuous improvement of automation and performance, the level of intelligence and automation is also constantly increasing, giving rise to various high-precision mechanical equipment. Affected by alternating loads and complex operating conditions, core power equipment inevitably experiences failures during long-term operation. Failures in high-end equipment exhibit characteristics such as complexity, subtle features, and uncertainty, making health status assessment and operation and maintenance of such devices extremely challenging. Currently, in industries such as manufacturing, communications, and energy, especially in complex industrial and information technology fields, delayed detection of equipment failures can easily lead to production interruptions, communication disruptions, or energy waste.

[0003] Furthermore, thanks to intelligent manufacturing and integrated manufacturing, various types of equipment are developing towards miniaturization, precision, and refinement. Simultaneously, fault monitoring, diagnosis, and early warning analysis of increasingly sophisticated and complex key components in various complex equipment have become crucial means to ensure the long-term safe operation of such equipment. This has garnered widespread attention from scholars, leading to the emergence of many efficient and highly accurate equipment operation and maintenance methods. However, accurately locating faults, especially those related to complex and vulnerable moving parts, after identifying equipment fault characteristics or early warning information, presents significant challenges. Utilizing limited sensor network resources for fault prediction, multi-fault isolation and location, and fault tracing in complex equipment has become an important research direction. Constructing a method for rapid backtracking, accurate location, and precise state prediction under multi-fault coupling is crucial for improving equipment reliability and is the core technical problem this patent aims to solve. Summary of the Invention

[0004] The purpose of this invention is to provide a method for extracting the evolution law of complex systems under multiple fault coupling conditions, which solves the problems of fault prediction and source tracing under coupled fault conditions.

[0005] The technical solution adopted in this invention is a method for extracting the evolution law of complex systems under multiple fault coupling conditions, which specifically includes the following steps:

[0006] S1. Construct a dataset of historical status information of the moving equipment and collect incremental data in real time;

[0007] S2. Preprocess the dataset and incremental data to obtain multiple sets of normalized data;

[0008] S3. Perform cluster analysis on multiple sets of normalized data to obtain clustering results and determine the number of categories N and the center of each category.

[0009] S4. By using the number of categories N and the center of each category, the data in S3 is backtracked to obtain multiple evolutionary clustering results;

[0010] S5. Extract patterns and preserve features from each clustering result to obtain multiple optimized evolutionary clustering results;

[0011] S6. Based on the optimized evolutionary clustering results, perform fault location and prediction on the newly added detection data;

[0012] S7. Based on the correlation between the predicted results and the measured data in S6, conduct a causal analysis.

[0013] The invention is further characterized in that:

[0014] The specific process of S1 is as follows: data is acquired in two forms: based on the existing database and on-site collection of sensor measurement points. Historical state information datasets and real-time incremental data are obtained at equal intervals. The measured time and characteristics are obtained, as well as the measured values ​​of different characteristics at each time point.

[0015] The specific process of S2 is as follows: outlier detection is performed on the dataset and incremental data. Outliers and noise are removed and missing values ​​are filled in using the 3δ criterion to obtain multiple sets of feature data. Then, the min-max method is used to normalize each set of feature data to obtain multiple sets of normalized data.

[0016] The specific method for removing outliers and noise and filling missing values ​​is as follows: For each feature, the data is tested separately, and the mean and standard deviation of each feature are calculated. Then, each data point for that feature is iterated through, and the absolute value of the difference between the data and the mean is calculated. It is then determined whether the difference is greater than three times the standard deviation. If it is greater than three times the standard deviation, the element is considered an outlier and replaced with the value of the preceding or following element. If it is less than or equal to three times the standard deviation, no action is taken.

[0017] The specific method for normalization using the maximum-minimum method is as follows: calculate the maximum and minimum values ​​of the data for each feature, and then normalize each feature, which is: (data - minimum value) / (maximum value - minimum value).

[0018] The specific process of S3 is as follows: Incremental clustering and k-means clustering are used to cluster multiple sets of normalized data to obtain clustering results, determine the number of categories N, obtain the category center of each category by calculating the average value of each feature, and indicate the category to which each time point belongs.

[0019] The specific process of S4 is as follows: By comparing the distances of the cluster item centers, each path cluster item can trace back to the historical state of the initial cluster item. All the cluster items on the same path form a fault backtracking path. Different evolution laws generate different fault backtracking paths, and the evolutionary clustering results of different paths are obtained through this process. The specific backtracking method is to start from the Nth category and trace back in the direction of the 1st category. Calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M (M < N). Then, start from the Mth category and continue to trace back forward until reaching the 1st category, forming the first backtracking path. Continue to perform path backtracking on the remaining categories in the same way to obtain the second backtracking path, and so on until all categories appear in the extracted paths.

[0020] The specific process of S5 is as follows: For each clustering result, the piecewise fuzzy statistical averaging method is adopted, and the order-preserving module of the software is used for order preservation. The temporal sequence of category evolution in each path is reorganized, and the samples that cause the historical states between clusters to cross are removed while retaining the situation with the most data on a single path, obtaining the optimized evolutionary clustering result.

[0021] The specific process of S6 is as follows: When a warning or alarm is caused by a fault sample point, a prediction method combining Gaussian interpolation and BiLSTM is used to analyze the sample data of multiple associated adjacent paths, that is, to predict and analyze the characteristic parameters of the optimized evolutionary clustering result in S5. The clustering analysis results of different paths are fused to挖掘 the internal evolution law, thereby judging the similarity degree of the key paths at different stages, that is, the probability size of each path corresponding to the fault. Based on this, fault analysis, prediction, and precise traceability are carried out.

[0022] The specific method is as follows: According to the software algorithm module, after model training, the prediction of future data and the probability size of each path corresponding to the fault are obtained. When there is a deviation from the predicted future sample set and the displayed sample state, fault location and prediction are realized.

[0023] The specific process of S7 is as follows: Based on the principle of reliefF for the data in S6, a software module for calculating importance is used to judge the specific location of the fault, analyze which specific features cause the fault, and obtain the fault cause.

[0024] The beneficial effects of the present invention are:

[0025] This invention provides a method for extracting the evolution patterns of complex systems under coupled fault conditions. Considering the coupled fault conditions, it utilizes a data-driven approach to conduct fault tracing, location, and prediction research, revealing the evolution patterns of different faults in the core moving parts of precision equipment. When processing data, this algorithm retains sequence information in addition to clustering for classification, enhancing its reliability. Furthermore, it offers strong real-time performance, enabling it to promptly meet the needs of scenarios requiring rapid response. It uncovers different fault evolution patterns hidden under the superposition of different fault modes, solving the problem of fault prediction and tracing under coupled fault conditions. Attached Figure Description

[0026] Figure 1 This is a deviation diagram between the predicted value and the actual value of backtracking path one in Embodiment 1 of the present invention;

[0027] Figure 2 This is a deviation diagram between the predicted value and the actual value of backtracking path two in Embodiment 1 of the present invention;

[0028] Figure 3 This is a deviation diagram between the predicted value and the actual value of backtracking path one in Embodiment 2 of the present invention;

[0029] Figure 4 This is a deviation diagram between the predicted value and the actual value of backtracking path two in Embodiment 2 of the present invention;

[0030] Figure 5 This is a deviation diagram between the predicted value and the actual value of backtracking path one in Embodiment 3 of the present invention;

[0031] Figure 6 This is a deviation diagram between the predicted value and the actual value of backtracking path two in Embodiment 3 of the present invention;

[0032] Figure 7 This is a deviation diagram between the predicted value and the actual value of backtracking path three in Embodiment 3 of the present invention. Detailed Implementation

[0033] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0034] The present invention provides a method for extracting the evolution law of complex systems under multiple fault coupling conditions, which specifically includes the following steps:

[0035] S1. Construct a dataset of historical status information of the moving equipment and collect incremental data in real time;

[0036] The specific process of S1 is as follows: data is acquired in two forms: based on the existing database and on-site collection of sensor measurement points. Historical status information datasets and real-time incremental data are obtained at equal intervals. The measured time and characteristics are obtained, as well as the measured values ​​of different characteristics at each time point.

[0037] S2. Preprocess the dataset and incremental data to obtain multiple groups of normalized data;

[0038] The specific process of S2 is as follows: Detect outliers in the dataset and incremental data, remove outliers and noise and fill in missing values through the 3δ criterion to obtain multiple groups of feature data, and perform normalization processing on each group of feature data using the maximum-minimum method to obtain multiple groups of normalized data;

[0039] The specific method for removing outliers and noise and filling in missing values is as follows: Detect the data of each feature separately, calculate the mean and standard deviation of each feature, traverse each data of the feature, calculate the absolute value of the difference between it and the mean, and determine whether it is greater than 3 times the standard deviation; if it is greater than 3 times the standard deviation, then this element is considered an outlier and is replaced with the value of the previous or next element; if it is less than or equal to 3 times the standard deviation, no processing is performed;

[0040] The specific method for normalization processing using the maximum-minimum method is as follows: Calculate the data of each feature separately, find the maximum and minimum values of the data under each feature, and then perform normalization calculation on each feature, that is: (this data - minimum value) / (maximum value - minimum value).

[0041] S3. Perform clustering analysis on multiple groups of normalized data to obtain clustering results, determine the number of categories N and the centers of each category;

[0042] The specific process of S3 is as follows: Use the methods of incremental clustering and k-means clustering to cluster multiple groups of normalized data to obtain clustering results, determine the number of categories N, obtain the category center of the corresponding category by calculating the average value of each feature in each category, and indicate the category to which each time point belongs.

[0043] S4. Through the number of categories N and the centers of each category, perform fault backtracking on the data in S3 to obtain multiple evolutionary clustering results;

[0044] The specific process of S4 is as follows: By comparing the distances between the cluster item centers, each path cluster item can be traced back to the historical state of the initial cluster item. All the cluster items on the same path form a fault backtracking path. Different evolutionary rules generate different fault backtracking paths. Different evolutionary clustering results of different paths are obtained through this process; the specific backtracking method is to start from the Nth category and trace back in the direction of the 1st category, calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M, M < N, then start from the Mth category and continue to trace back forward until tracing back to the 1st category to form the 1st backtracking path. Continue to perform path backtracking on the remaining categories in the same way to obtain the 2nd backtracking path, and so on until all categories appear in the extracted paths.

[0045] S5. Extract patterns and preserve features from each clustering result to obtain multiple optimized evolutionary clustering results;

[0046] The specific process of S5 is as follows: each clustering result is subjected to a segmented fuzzy statistical averaging method, and the order preservation module of the software is used to preserve the order. The evolution time sequence of the categories in each path is reorganized, and samples that cause the historical states between clusters are removed while retaining the most data on a single path, so as to obtain the optimized evolutionary clustering results.

[0047] S6. Based on the optimized evolutionary clustering results, perform fault location and prediction on the newly added detection data;

[0048] The specific process of S6 is as follows: When a warning or alarm is triggered by a fault sample point, a prediction method based on Gaussian interpolation and BiLSTM is used to predict and analyze the feature parameters of multiple related and adjacent path sample data, that is, the optimized evolutionary clustering results in S5. The clustering analysis results of different paths are fused to explore the inherent evolutionary rules, thereby judging the similarity of critical paths at different stages, that is, the probability of each path corresponding to the fault. Based on this, the fault is analyzed, predicted, and accurately traced.

[0049] The specific method is as follows: after the software algorithm module has been trained, it calculates the prediction of future data and the probability of each path corresponding to the fault; when there is a deviation between the predicted future sample set and the displayed sample status, the fault location and prediction are realized.

[0050] S7. Based on the correlation between the predicted results and the measured data in S6, conduct a causal analysis;

[0051] The specific process of S7 is as follows: Based on the reliefF principle, the software module for importance calculation is used to determine the specific location of the fault, analyze which specific characteristics caused the fault, and obtain the cause of the fault.

[0052] Example 1

[0053] The method for extracting the evolution law of complex systems under multi-fault coupled conditions in this embodiment specifically includes the following steps:

[0054] S1. Construct a dataset of historical status information of the moving equipment and collect incremental data in real time;

[0055] Data was acquired using two methods: existing databases and on-site data collection from sensor measurement points. Historical state information datasets and real-time incremental data were obtained at equal intervals. The measured time and characteristics, as well as the measured values ​​of different characteristics at each time point, are shown in Table 1.

[0056] Table 1 Excerpt of Data Collected and Analyzed on-site

[0057] -188.4163 579.3932 45.1773 1399.6958 -185.5324 -188.4163 578.8117 45.1949 1399.1711 -185.5222 -188.4163 580.2703 45.2375 1398.8418 -185.4843 ... ... ... ... ... -188.3842 578.1443 45.2059 1398.71 -185.4956 -188.3842 577.9713 45.185 1400.5132 -185.4958 -188.3842 578.2301 45.2107 1400.9609 -185.4961 -188.3842 577.6804 45.2424 1401.5283 -185.4843

[0058] S2. Preprocess the dataset and incremental data to obtain multiple sets of normalized data;

[0059] The specific process is as follows: outlier detection is performed on the dataset and incremental data, outliers and noise are removed and missing values ​​are filled in using the 3δ criterion to obtain multiple sets of feature data, and the min-max method is used to normalize each set of feature data to obtain multiple sets of normalized data.

[0060] The specific method for removing outliers and noise and filling missing values ​​is as follows: For each feature, the data is tested separately, and the mean and standard deviation of each feature are calculated. Then, each data point for that feature is iterated through, and the absolute value of the difference between the data and the mean is calculated. It is then determined whether the difference is greater than three times the standard deviation. If it is greater than three times the standard deviation, the element is considered an outlier and replaced with the value of the preceding or following element. If it is less than or equal to three times the standard deviation, no action is taken.

[0061] The specific method for normalization using the max-min method is as follows: Calculate the maximum and minimum values ​​of the data for each feature, and then normalize each feature using the formula: (data value - minimum value) / (maximum value - minimum value), as shown in Table 2.

[0062] Table 2 shows the preprocessed data obtained after step S2.

[0063] 0 0.661338276 0.257964258 0.349785332 0 0 0.436812232 0.394716395 0.163609268 0.212058212 0 1 0.725718726 0.046765781 1 ... ... ... ... ... 1 0.179118885 0.48018648 0 0.765072765 1 0.112320939 0.317793318 0.63981833 0.760914761 1 0.212247577 0.517482517 0.798672959 0.754677755 1 0 0.763791764 1 1

[0064] S3. Perform cluster analysis on multiple sets of normalized data to obtain clustering results and determine the number of categories N and the center of each category.

[0065] The specific process is as follows: Incremental clustering and k-means clustering are used to cluster multiple sets of normalized data to obtain clustering results. The number of categories N is determined. In each category, the category center is obtained by calculating the average value of each feature, and the category to which each time point belongs is indicated, as shown in Table 3.

[0066] Table 3 shows the clustering results obtained after step S3.

[0067]

[0068] S4. By using the number of categories N and the center of each category, the data in S3 is backtracked to obtain multiple evolutionary clustering results;

[0069] The specific process is as follows: By comparing the distances between the cluster item centers, each path of cluster items can trace back to the historical state of the initial cluster item. All the cluster items on the same path form a fault traceback path. Different evolution laws generate different fault traceback paths. Through this process, the evolutionary clustering results of different paths are obtained. The specific traceback method is to start from the Nth category and trace back in the direction of the 1st category. Calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M, where M < N. Then, starting from the Mth category, continue to trace back forward until reaching the 1st category, forming the 1st traceback path. For the remaining categories, continue the path traceback in the same way to obtain the 2nd traceback path, and so on until all categories appear in the extracted paths, as shown in Table 4:

[0070] Table 4 Traceback paths obtained after the S4 step

[0071] 4 3 1 0 2 1 0 0

[0072] S5. Extract the laws and retain the features of each clustering result to obtain multiple optimized evolutionary clustering results;

[0073] The specific process is as follows: For each clustering result, use the piecewise fuzzy statistical averaging method and the order-preserving module of the software to preserve the order, reorganize the time sequence of category evolution in each path, and remove the samples that cause the historical states between clusters to intersect while retaining the situation with the most data on a single path, obtaining the optimized evolutionary clustering results (the last category represents the category to which it belongs), as shown in Tables 5 and 6:

[0074] Table 5 Order-preserving results for the first traceback path obtained after the S5 step

[0075] 1 2 3 4 5 6 0.775700935 1 1 1 1 1 ... ... ... ... ... ... 0.215484512 0.467196537 0 1 0.63981833 0.798672959 1 0 0.765072765 1 0.760914761 0.754677755 1 1 1 3 4 4

[0076] Table 6 Order-preserving results for the second traceback path obtained after the S5 step

[0077]

[0078]

[0079] S6. Based on the optimized evolutionary clustering results of each item, perform fault location and prediction on the newly added detection data;

[0080] The specific process is as follows: When a warning or alarm is triggered by a fault sample point, a prediction method based on Gaussian interpolation and BiLSTM is used to predict and analyze the feature parameters of multiple related adjacent path sample data, that is, the optimized evolutionary clustering results in S5. The clustering analysis results of different paths are fused to explore the inherent evolutionary rules, thereby judging the similarity of critical paths at different stages, that is, the probability of each path corresponding to the fault. Based on this, the fault is analyzed, predicted, and accurately traced.

[0081] The specific method is as follows: Based on the software algorithm module's model training, its predictions of future data and the probability of each path corresponding to the fault are calculated. When the deviation from the predicted future sample set and the displayed sample state exceeds a set threshold, fault location and prediction are achieved. Figure 1 , 2 As shown;

[0082] S7. Based on the correlation between the predicted results and the measured data in S6, conduct a causal analysis;

[0083] The two paths obtained in S4 are subjected to the order preservation step in S5 and the LSTM prediction step in S6 respectively to obtain the predicted values ​​at the same time node. By comparing the similarity between the predicted values ​​and the actual values ​​of the two paths, the prediction similarity of the first path, i.e., 4-3-1, is the highest, which shows that the current state evolved from the first path.

[0084] Example 2

[0085] The method for extracting the evolution law of complex systems under multi-fault coupled conditions in this embodiment specifically includes the following steps:

[0086] S1. Construct a dataset of historical status information of the moving equipment and collect incremental data in real time;

[0087] Data was acquired using two methods: existing databases and on-site data collection from sensor measurement points. Historical state information datasets and real-time incremental data were obtained at equal intervals. The measured time and characteristics, as well as the measured values ​​of different characteristics at each time point, are shown in Table 7.

[0088] Table 7 Excerpt of Data Collected and Analyzed on-site

[0089] -188.3842 579.4174 45.3368 1400.65 -185.4843 -188.3842 579.8708 45.3816 1400.9722 -185.4843 -188.3842 580.6398 45.3132 1400.535 -185.4843 ... ... ... ... ... -188.3842 579.0989 45.4245 1399.7368 -185.4843 -188.3842 578.6845 45.3797 1400.3741 -185.4843 -188.3842 580.5079 45.4232 1398.8256 -185.4843 -188.3362 579.319 45.4644 1399.6002 -185.4843

[0090] S2. Preprocess the dataset and incremental data to obtain multiple sets of normalized data;

[0091] The specific process is as follows: outlier detection is performed on the dataset and incremental data, outliers and noise are removed and missing values ​​are filled in using the 3δ criterion to obtain multiple sets of feature data, and the min-max method is used to normalize each set of feature data to obtain multiple sets of normalized data.

[0092] The specific method for removing outliers and noise and filling missing values ​​is as follows: For each feature, the data is tested separately, and the mean and standard deviation of each feature are calculated. Then, each data point for that feature is iterated through, and the absolute value of the difference between the data and the mean is calculated. It is then determined whether the difference is greater than three times the standard deviation. If it is greater than three times the standard deviation, the element is considered an outlier and replaced with the value of the preceding or following element. If it is less than or equal to three times the standard deviation, no action is taken.

[0093] The specific method for normalization using the max-min method is as follows: Calculate the maximum and minimum values ​​of the data for each feature, and then normalize each feature using the formula: (data value - minimum value) / (maximum value - minimum value), as shown in Table 8.

[0094] Table 8 shows the preprocessed data obtained after step S2.

[0095] 0.306358382 0.374827392 0.156084656 0.873884453 0.5 0.306358382 0.606709968 0.452380952 1 0.5 0.306358382 1 0 0.828871145 0.5 ... ... ... ... ... 0.306358382 0.211936787 0.736111111 0.516439643 0.5 0.306358382 0 0.439814815 0.765891655 0.5 0.306358382 0.932542321 0.727513228 0.159777673 0.5 1 0.324502634 1 0.462971661 0.5

[0096] S3. Perform cluster analysis on multiple sets of normalized data to obtain clustering results and determine the number of categories N and the center of each category.

[0097] The specific process is as follows: Incremental clustering and k-means clustering are used to cluster multiple sets of normalized data to obtain clustering results. The number of categories N is determined. In each category, the category center is obtained by calculating the average value of each feature. The category to which each time point belongs is indicated (the last row indicates the category), as shown in Table 9.

[0098] Table 9 shows the clustering results obtained after step S3.

[0099]

[0100] S4. By using the number of categories N and the center of each category, the data in S3 is backtracked to obtain multiple evolutionary clustering results;

[0101] The specific process is as follows: By comparing the distances of the cluster item centers, each path's cluster items can trace back to the historical state of the initial cluster item. All the cluster items on the same path form a fault traceback path. Different evolution laws generate different fault traceback paths. Through this process, the evolutionary clustering results of different paths are obtained. The specific traceback method is to start from the Nth category and trace back in the direction of the 1st category. Calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M, where M < N. Then, starting from the Mth category, continue to trace back forward until reaching the 1st category, forming the 1st traceback path. For the remaining categories, continue the path traceback in the same way to obtain the 2nd traceback path, and so on until all categories appear in the extracted paths, as shown in Table 10:

[0102] Table 10 Traceback paths obtained after the S4 step

[0103] 5 4 3 1 0 2 1 0 0 0

[0104] S5. Extract the laws and retain the features of each clustering result to obtain multiple optimized evolutionary clustering results;

[0105] The specific process is as follows: For each clustering result, use the piecewise fuzzy statistical averaging method and the order-preserving module of the software to preserve the order, reorganize the temporal sequence of category evolution in each path, and remove the samples that cause the historical states between clusters to cross while retaining the situation with the most data on a single path, obtaining the optimized evolutionary clustering results, as shown in Tables 11 and 12:

[0106] Table 11 Order-preserving results of traceback path one obtained after the S5 step

[0107]

[0108] Table 12 Order-preserving results of traceback path two obtained after the S5 step

[0109]

[0110]

[0111] S6. Based on the optimized evolutionary clustering results of each path, perform fault location and prediction on the newly added detection data;

[0112] The specific process is as follows: When a warning or alarm is triggered by a fault sample point, a prediction method based on Gaussian interpolation and BiLSTM is used to predict and analyze the feature parameters of multiple related adjacent path sample data, that is, the optimized evolutionary clustering results in S5. The clustering analysis results of different paths are fused to explore the inherent evolutionary rules, thereby judging the similarity of critical paths at different stages, that is, the probability of each path corresponding to the fault. Based on this, the fault is analyzed, predicted, and accurately traced.

[0113] The specific method is as follows: Based on the software algorithm module's model training, its predictions of future data and the probability of each path corresponding to the fault are obtained; when there is a deviation between the predicted future sample set and the displayed sample state, fault location and prediction are achieved, such as... Figure 3 , 4 As shown;

[0114] S7. Based on the correlation between the predicted results and the measured data in S6, conduct a causal analysis;

[0115] The two paths obtained in S4 are subjected to the order-preserving step in S5 and the LSTM prediction step in S6, respectively, to obtain predicted values ​​for the same time node. By comparing the similarity between the predicted values ​​and the actual values ​​of the two paths, the first path, 5-4-3-1, has the highest prediction similarity. This indicates that the current state evolved from the first path.

[0116] Example 3

[0117] The method for extracting the evolution law of complex systems under multi-fault coupled conditions in this embodiment specifically includes the following steps:

[0118] S1. Construct a dataset of historical status information of the moving equipment and collect incremental data in real time;

[0119] Data was acquired using two methods: existing databases and on-site data collection from sensor measurement points. Historical state information datasets and real-time incremental data were obtained at equal intervals. The measured time and characteristics, as well as the measured values ​​of different characteristics at each time point, are shown in Table 13.

[0120] Table 13 Excerpt of Data Collected and Analyzed on-site

[0121] -188.3362 580.4955 45.4736 1400.9868 -185.4843 -188.3362 581.6773 45.5152 1400.3488 -185.4843 -188.3595 581.3552 45.5404 1399.8976 -185.4843 ... ... ... ... ... -188.3362 582.1576 45.5449 1400.8105 -185.4685 -188.3197 580.9806 45.6458 1400.4916 -185.4843 -188.3195 581.8618 45.7178 1400.6707 -185.4689 -188.3362 582.7071 45.682 1401.0012 -185.4675

[0122] S2. Preprocess the dataset and incremental data to obtain multiple sets of normalized data;

[0123] The specific process is as follows: outlier detection is performed on the dataset and incremental data, outliers and noise are removed and missing values ​​are filled in using the 3δ criterion to obtain multiple sets of feature data, and the min-max method is used to normalize each set of feature data to obtain multiple sets of normalized data.

[0124] The specific method for removing outliers and noise and filling missing values ​​is as follows: For each feature, the data is tested separately, and the mean and standard deviation of each feature are calculated. Then, each data point for that feature is iterated through, and the absolute value of the difference between the data and the mean is calculated. It is then determined whether the difference is greater than three times the standard deviation. If it is greater than three times the standard deviation, the element is considered an outlier and replaced with the value of the preceding or following element. If it is less than or equal to three times the standard deviation, no action is taken.

[0125] The specific method for normalization using the max-min method is as follows: Calculate the maximum and minimum values ​​of the data for each feature, and then normalize each feature using the formula: (data value - minimum value) / (maximum value - minimum value), as shown in Table 14.

[0126] Table 14 shows the preprocessed data obtained after step S2.

[0127] 0.5825 0 0 0.986951794 0 0.5825 0.534364261 0.17035217 0.408843784 0 0 0.388723096 0.273546274 0 0 ... ... ... ... ... 0.5825 0.751537349 0.291973792 0.827201885 0.49375 0.995 0.219343462 0.705159705 0.538238492 0 1 0.617788027 1 0.700525553 0.48125 0.5825 1 0.853398853 1 0.525

[0128] S3. Perform cluster analysis on multiple sets of normalized data to obtain clustering results and determine the number of categories N and the center of each category.

[0129] The specific process is as follows: Incremental clustering and k-means clustering are used to cluster multiple sets of normalized data to obtain clustering results. The number of categories N is determined. In each category, the category center is obtained by calculating the average value of each feature, and the category to which each time point belongs is indicated, as shown in Table 15.

[0130] Table 15 shows the clustering results obtained after step S3.

[0131] 0.30635 0.30635 0.30635 0 0.30635 0.30635 0.30635 0.30635 0.30635 1 0.37482 0.60671 1 0.61709 0.75548 0.72311 0.21193 0 0.93254 0.3245 ... ... ... ... ... ... ... ... ... ... 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 4 4 4 2 3 4 3 3 2 1

[0132] S4. By using the number of categories N and the center of each category, the data in S3 is backtracked to obtain multiple evolutionary clustering results;

[0133] The specific process is as follows: By comparing the distances of the clustering item centers, each path clustering item can trace back to the historical state of the clustering item at the beginning. All the clustering items on the same path form a fault backtracking path. Different evolution laws generate different fault backtracking paths. Through this process, the evolutionary clustering results of different paths are obtained. The specific backtracking method is to start from the Nth category and trace back in the direction of the 1st category. Calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M, where M < N. Then, starting from the Mth category, continue to trace back forward until reaching the 1st category, forming the 1st backtracking path. Continue to perform path backtracking on the remaining categories in the same way to obtain the 2nd backtracking path, and so on until all categories appear in the extracted paths, as shown in Table 16:

[0134] Table 16 Backtracking paths obtained after the S4 step

[0135] 4 1 0 0 3 1 0 0 2 1 0 0

[0136] S5. Extract the rules and retain the features of each clustering result to obtain multiple optimized evolutionary clustering results;

[0137] The specific process is as follows: For each clustering result, use the piecewise fuzzy statistical averaging method and the order-preserving module of the software to perform order-preserving, reorganize the time sequence of category evolution in each path, and remove the samples that cause the historical states between clusters to cross while retaining the situation with the most data on a single path, obtaining the optimized evolutionary clustering results, as shown in Tables 17 - 19:

[0138] Table 17 Order-preserving results for the first backtracking path obtained after the S5 step

[0139]

[0140] Table 18 Order-preserving results for the second backtracking path obtained after the S5 step

[0141]

[0142]

[0143] Table 19 Order-preserving results for the third backtracking path obtained after the S5 step

[0144]

[0145] S6. Based on the optimized evolutionary clustering results of each path, perform fault location and prediction on the newly added detection data;

[0146] The specific process is as follows: When a warning or alarm is triggered by a fault sample point, a prediction method based on Gaussian interpolation and BiLSTM is used to predict and analyze the feature parameters of multiple related adjacent path sample data, that is, the optimized evolutionary clustering results in S5. The clustering analysis results of different paths are fused to explore the inherent evolutionary rules, thereby judging the similarity of critical paths at different stages, that is, the probability of each path corresponding to the fault. Based on this, the fault is analyzed, predicted, and accurately traced.

[0147] The specific method is as follows: Based on the software algorithm module's model training, its predictions of future data and the probability of each path corresponding to the fault are obtained; when there is a deviation between the predicted future sample set and the displayed sample state, fault location and prediction are achieved, such as... Figure 5-7 As shown;

[0148] S7. Based on the correlation between the predicted results and the measured data in S6, conduct a causal analysis;

[0149] The three paths obtained from S4 are subjected to the order-preserving step in S5 and the LSTM prediction step in S6 respectively to obtain the predicted values ​​for the same time node. By comparing the similarity between the predicted values ​​and the actual values ​​of the three paths, the prediction similarity of the first path, i.e., 4-1, is the highest, which shows that the current state evolved from the first path.

Claims

1. A method for extracting the evolution law of complex systems under multi-fault coupled conditions, characterized in that, Specifically, it includes the following steps: S1. Construct a historical state information dataset of dynamic equipment and collect incremental data in real time; S2. Preprocess the dataset and incremental data to obtain multiple groups of normalized data; S3. Conduct cluster analysis on multiple groups of normalized data to obtain a clustering result, determine the number of categories N and the center of each category; S4. Through the number of categories N and the center of each category, perform fault backtracking on the data in S3 to obtain multiple evolutionary clustering results; S5. Extract rules and retain features for each clustering result to obtain multiple optimized evolutionary clustering results; S6. Based on each optimized evolutionary clustering result, perform fault location and prediction on newly detected data; S7. Conduct cause analysis based on the correlation between the prediction result in S6 and the measured data; The specific process of S4 is as follows: By comparing the distances between the centers of clustering items, each path of clustering items can be traced back to the clustering item with the earliest historical state. All clustering items on the same path form a fault backtracking path, and different evolutionary laws produce different fault backtracking paths. In this process, evolutionary clustering results of different paths are obtained. The specific backtracking method is to start from the Nth category and trace back in the direction of the 1st category. Calculate the inter-class similarity between the Nth category and other categories to determine the most similar category M, where M < N. Then, start from the Mth category and continue to trace back forward until reaching the 1st category, forming the 1st backtracking path. Continue to perform path backtracking on the remaining categories in the same way to obtain the 2nd backtracking path, and so on until all categories appear in the extracted paths.

2. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S1 is as follows: Obtain data through two forms, namely, based on the existing database and on-site collection of sensor measurement point distributions, and obtain the historical state information dataset and real-time incremental data at equal intervals, obtaining the measured time and features, as well as the values measured for different features at each time point.

3. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S2 is as follows: Perform outlier detection on the dataset and incremental data,剔除异常值及噪声并填补缺失值,得到多组特征数据,并对每组特征数据使用最大最小法进行归一化处理,得到多组归一化数据。 4. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 3, characterized in that, The specific method for剔除异常值及噪声并填补缺失值 is as follows: Detect the data of each feature separately, calculate the mean and standard deviation of each feature, traverse each data of the feature, calculate the absolute value of the difference between it and the mean, and determine whether it is greater than 3 times the standard deviation. If it is greater than 3 times the standard deviation, then consider this feature as an outlier and replace it with the value of the previous or next feature. If it is less than or equal to 3 times the standard deviation, no processing is performed; The specific method for normalizing using the maximum-minimum method is as follows: Calculate the data of each feature separately, find the maximum and minimum values of the data under each feature, and then perform normalized calculation for each feature, that is: (this data - minimum value) / (maximum value - minimum value). It should be noted that there seems to be some text missing or incorrect in the original Chinese for item . The translation is done based on the existing text as accurately as possible.

5. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S3 is as follows: Incremental clustering and k-means clustering are used to cluster multiple sets of normalized data to obtain clustering results, determine the number of categories N, obtain the category center of each category by calculating the average value of each feature, and indicate the category to which each time point belongs.

6. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S5 is as follows: each clustering result is subjected to a segmented fuzzy statistical averaging method, and the order preservation module of the software is used to preserve the order. The evolution time sequence of the categories in each path is reorganized, and samples that cause the historical states between clusters to overlap are removed while retaining the most data on a single path, so as to obtain the optimized evolutionary clustering result.

7. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S6 is as follows: When a warning or alarm is triggered by a fault sample point, a prediction method based on Gaussian interpolation and BiLSTM is used to predict and analyze the feature parameters of multiple related adjacent path sample data, that is, the optimized evolutionary clustering results in S5. The clustering analysis results of different paths are fused to explore the inherent evolutionary rules, thereby judging the similarity of critical paths at different stages, that is, the probability of each path corresponding to the fault. Based on this, the fault is analyzed, predicted, and accurately traced. The specific method is as follows: after the software algorithm module has been trained, it calculates the prediction of future data and the probability of each path corresponding to the fault; when there is a deviation between the predicted future sample set and the displayed sample state, the fault location and prediction are realized.

8. The method for extracting the evolution law of complex systems under multi-fault coupled conditions according to claim 1, characterized in that, The specific process of S7 is as follows: based on the reliefF principle, the software module for importance calculation is used to determine the specific location of the fault, analyze which specific characteristics caused the fault, and obtain the cause of the fault.

Images