An information-based intelligent operation and maintenance method and system

Abstract

The present invention relates to the field of data processing, and particularly to an information-based intelligent operation and maintenance method and system. By acquiring key operation parameters of devices and performing data preprocessing, secondly, dimensionality reduction is optimized for device operation state classification based on the LDA algorithm, and at the same time, a feature weighting mechanism is introduced based on the correlation of operation parameters of devices in different operation states to improve the discrimination of the data after dimensionality reduction and the accuracy of operation and maintenance decisions. Finally, the information-based intelligent operation and maintenance of devices is realized by combining machine learning models and visualization technologies, etc. By adopting LDA dimensionality reduction and feature weighting optimization, and combining machine learning analysis and an intelligent operation and maintenance system, the present invention improves the accuracy of device state classification, fault prediction and anomaly detection, and realizes efficient intelligent operation and maintenance.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and particularly to an information-based intelligent operation and maintenance method and system. Background Art

[0002] In a modern enterprise environment, various computers, servers, and other IT devices constitute the infrastructure of the enterprise information system. With the expansion of business, the number of devices has increased sharply, and enterprises need efficient operation and maintenance management means to ensure the stable operation of devices. However, traditional operation and maintenance modes rely on manual inspections, log analysis, etc., with low efficiency and poor real-time performance, unable to meet the requirements of complex IT environments. Especially in large-scale data centers, cloud computing platforms, and enterprise internal server groups, equipment failures, resource utilization optimization, and performance monitoring have become core operation and maintenance challenges.

[0003] With the increase in the types of devices and monitoring data, enterprises face multiple technical problems: such as data redundancy and noise, where a large amount of monitoring data contains duplicate, irrelevant, or incorrect data, affecting the analysis accuracy. The high-dimensional complexity of data; and the operation of devices involves multi-dimensional features, and direct analysis has the problem of dimensionality disaster, increasing the computational complexity; it also includes the lag of traditional operation and maintenance methods: faults are usually responded to after the event, and it is difficult to predict potential problems in a timely manner, resulting in operation and maintenance lag and high maintenance costs. In response to the above problems, there are some traditional dimensionality reduction methods for optimization. In the existing technology of data dimensionality reduction, LDA (Linear Discriminant Analysis) is a commonly used dimensionality reduction method.

[0004] LDA maps data from a high-dimensional space to a low-dimensional space by maximizing the ratio of between-class scatter to within-class scatter. However, in actual application scenarios, LDA cannot effectively handle the correlation between features in the calculation of the within-class scatter matrix, resulting in the data space after dimensionality reduction may not fully reflect the true state of the device. This defect may cause the intelligent operation and maintenance system to fail to accurately identify potential device anomalies in tasks such as fault prediction, resource scheduling, and alarm, affecting the accuracy of decision-making. Summary of the Invention

[0005] In view of the above problem of being unable to effectively handle the correlation between features, in the first aspect, the present invention proposes an information-based intelligent operation and maintenance method, including: collecting the operation parameters of multiple devices and calculating the operation fluency, and classifying each device based on the operation fluency; the operation parameters of all devices constitute a data set, and the data set is dimensionally reduced using the LDA algorithm based on the category to obtain the dimensionally reduced data; clustering the data to obtain a clustering result, and judging whether there is an abnormality in the device based on the clustering result; the LDA algorithm also includes processing the within-class scatter matrix to obtain a feature-weighted within-class scatter matrix, and the feature-weighted within-class scatter matrix is the product of the feature weight matrix and each feature vector in the within-class scatter matrix; the calculation method of the feature weight matrix is specifically: dividing the data set into multiple classification data sets according to the device category, sorting the same operation parameter of all devices in the classification data set according to the device number to obtain a corresponding feature parameter sequence; calculating the ratio of the variance to the mean of the feature parameter sequence to obtain the concentration; calculating the absolute value of the difference between the maximum value and the minimum value in each feature parameter sequence; for any two feature parameter sequences, taking the ratio of twice the corresponding absolute value minimum to the sum of the corresponding absolute values as the synchronous volatility; and calculating the ratio of the synchronous volatility to the corresponding DTW distance to obtain the co-variation index; taking the ratio of the sum of the co-variation indices of each feature parameter sequence and the remaining feature parameter sequences to the corresponding concentration as the weight of the corresponding feature, and the weights of all features constitute the feature weight matrix.

[0006] The present invention combines the operation fluency of devices for data classification and dimensionality reduction, significantly improving the accuracy and efficiency of the information-based intelligent operation and maintenance method. Compared with the prior art, traditional operation and maintenance methods often rely on single features or simple statistical indicators to evaluate the operation status of devices, while the present invention comprehensively considers the operation parameters of multiple devices, calculates the operation fluency of devices, and classifies devices based on the fluency. Especially in the LDA dimensionality reduction process, the introduction of the feature-weighted within-class scatter matrix further enhances the sensitivity of the algorithm to the relationship between features, making the dimensionally reduced data more concise and having a higher class discrimination. Further identifying whether there is an abnormality in the device through the clustering algorithm can more efficiently and accurately warn of faults. This method breaks through the limitations of traditional operation and maintenance methods, can process complex device data, and improves the accuracy and response speed of device status recognition in the dimensionality reduction and clustering processes, and is particularly suitable for intelligent operation and maintenance environments where device status changes frequently and complexly.

[0007] Further, the calculation method of the feature-weighted within-class scatter matrix is specifically:

[0008] ;

[0009] where S * Wdenotes the feature-weighted within-class scatter matrix; k denotes the total number of classes; denotes the class dataset; denotes the feature parameter weight matrix of the class dataset; X denotes the feature vector of any device in the denotes the mean vector of the

[0010] Through the calculation of the feature-weighted within-class scatter matrix, the process of LDA dimensionality reduction is further refined. Compared with the traditional LDA algorithm that only relies on the within-class scatter matrix, this method takes into account the weighted influence of device features, thereby enhancing the classification effect of the dimensionality reduction process. This method enables the features of devices to be more accurately represented according to class differences, improving the device status recognition ability of the intelligent operation and maintenance system and the accuracy of anomaly warning.

[0011] Furthermore, the specific calculation method of the weight is as follows:

[0012] ;

[0013] where denotes the weight of the j-th feature parameter in the class dataset; denotes the co-variability index of the remaining feature parameter sequences and the j-th feature parameter sequence in the j class dataset; H denotes the concentration of the j-th feature parameter sequence in the

[0014] By designing the calculation method of feature parameter weights, the classification accuracy of devices and the adaptive ability of the system are further optimized. Compared with the prior art, the present invention dynamically evaluates the contribution degree of each feature to the device status judgment by combining the co-variability index and the concentration, solving the defect of relying on fixed weight values in the traditional method. This method can flexibly adapt to the feature differences between different device classes, improving the accurate evaluation and warning ability of the operation and maintenance system for the device operation status.

[0015] Furthermore, the specific calculation method of the co-variability index is as follows:

[0016] ;

[0017] Among them, U(j, p) represents the covariance index between the j-th feature parameter sequence and the p-th feature parameter sequence; R(j, p) represents the synchronous volatility between the j-th feature parameter sequence and the p-th feature parameter sequence; Dtw(j, p) represents the DTW distance between the j-th feature parameter sequence and the p-th feature parameter sequence; σ represents a tiny constant.

[0018] By introducing the calculation method of the covariance index, the mutual relationship between feature parameter sequences can be measured more precisely. Compared with the traditional covariance method, the present invention calculates the covariance between features by combining the comprehensive index of synchronous volatility and DTW distance, can effectively handle the multi-dimensional dependence relationship of the device in a complex operating environment, thereby improving the dimensionality reduction effect of the LDA algorithm in high-dimensional data and enhancing the accuracy of device status analysis.

[0019] Further, the calculation method of the synchronous volatility is specifically as follows:

[0020] ;

[0021] Among them, R(j, p) represents the synchronous volatility between the j-th feature parameter sequence and the p-th feature parameter sequence; max( ) and min( ) respectively represent the functions of taking the maximum value and the minimum value; represents the j-th feature parameter sequence; represents the p-th feature parameter sequence.

[0022] Through the calculation method of synchronous volatility, the similarity measurement between feature sequences is further strengthened. In traditional technologies, the calculation of the correlation between features often ignores the importance of fluctuation consistency in evaluating the operating state of the device. By taking synchronous volatility as a new measurement criterion, the present invention can consider the fluctuation consistency of feature sequences during device status classification, thereby more accurately reflecting the operating condition of the device and effectively avoiding the misjudgment problem that may occur in traditional methods.

[0023] Further, the calculation method of the operation smoothness is specifically as follows:

[0024] ;

[0025] Among them, F i represents the operation smoothness of the i-th device; exp( ) represents the exponential function; α represents the adjustment factor; (X i,1 ) represents the CPU usage rate; log( ) represents the logarithmic function; (X i,3 ) represents the memory usage rate; represents a tiny constant; represents the hard disk read / write speed (X i,5 ) and the response time (X i,7)'s relative relationship; γ represents a tiny constant.

[0026] Further, based on the classified running fluency, the categories of each device are obtained, and it further includes: in response to the running fluency of the device being greater than or equal to the fluency threshold, classifying the device as running smoothly; in response to the running fluency of the device being greater than or equal to the lag threshold and less than the fluency threshold, classifying the device as running relatively smoothly; in response to the running fluency of the device being less than the lag threshold, classifying the device as running lagging.

[0027] Further, it also includes preprocessing the running parameters of the collected multiple devices, specifically: removing duplicate running parameters by detecting the uniqueness of the timestamp and device ID; filling in missing values of the running parameters using the interpolation method of adjacent time points; standardizing the running parameters using the Z-score algorithm; removing short-term noise from the running parameters through time window sliding average processing.

[0028] Further, clustering the data to obtain a clustering result, and based on the clustering result, determining whether the device is abnormal, and it further includes: using the K-Means algorithm to cluster the data to obtain multiple clustering clusters; calculating the Euclidean distance between each data and the central data of the corresponding clustering cluster; in response to the Euclidean distance being greater than the set distance threshold, determining that the corresponding device is abnormal and notifying the operation and maintenance personnel to confirm the status.

[0029] In a second aspect, the present invention provides an information-based intelligent operation and maintenance system, including a processor and a memory, and the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the information-based intelligent operation and maintenance method of the present invention is implemented.

[0030] The technical effects of the present invention are as follows:

[0031] The present invention calculates the running fluency of the device by comprehensively considering various device running parameters, classifies the device based on the fluency, and combines the LDA algorithm for data dimensionality reduction processing, realizing efficient device status evaluation and anomaly detection. In particular, a calculation method of the feature-weighted within-class scatter matrix is proposed, and through the analysis of the custom parameter sequence characteristics and the synchronous volatility measurement, the correlation between features is accurately measured, improving the accuracy of dimensionality reduction and classification. In addition, the present invention uses a unique clustering algorithm combined with the Euclidean distance to detect device anomalies, enhancing the sensitivity of the intelligent operation and maintenance system to device status changes, having high adaptability and accuracy, and providing an innovative solution for device management in complex environments. Brief Description of the Drawings

[0032] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown by way of example and not limitation, and like or corresponding reference numerals indicate like or corresponding parts, wherein:

[0033] Figure 1 is a flowchart of an information-based intelligent operation and maintenance method for an embodiment in the present invention shown schematically;

[0034] Figure 2 is a block diagram of an information-based intelligent operation and maintenance system structure for an embodiment in the present invention shown schematically. Specific Embodiments

[0035] Next, the technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] Next, the specific embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0037] Embodiment of the Information-based Intelligent Operation and Maintenance Method:

[0038] As Figure 1 shown, the information-based intelligent operation and maintenance method of the present invention includes:

[0039] S1. Collect device operation and maintenance data and perform data preprocessing; calculate the smoothness of device operation and classify to obtain a data set.

[0040] In the process of information-based intelligent operation and maintenance, the operation and maintenance data of devices is the basis for realizing intelligent decision-making and optimized scheduling. With the increase in the types of devices and the expansion of data volume, the data processing problems faced by enterprises are becoming increasingly serious. Especially in the multi-dimensional and multi-level features of device status and fault information, redundant information and noise seriously affect the analysis effect, resulting in delays and inaccuracies in operation and maintenance decisions. Therefore, in this embodiment, dimensionality reduction technology can be used to process the data, effectively simplify the data structure, remove irrelevant or redundant parts, and retain the most important feature information to support the subsequent operation and maintenance management and decision-making process.

[0041] S1.1. Collect device operation and maintenance data and perform data preprocessing.

[0042] First, the CPU usage rate and memory usage rate can be obtained through the API interfaces provided by various devices; then, performance detection tools such as Nagios can be used to collect the load balance degree, memory bandwidth, response time, disk read and write speed, and network traffic of the devices; finally, temperature sensors such as LM75 can be used to collect the temperatures of various devices. It should be noted that: in this embodiment, the collection frequency of all data can be set to 5 times / min. In summary, the relevant data obtained for the i-th device at the same moment includes: CPU usage rate (X i,1 ), in percentage; load balance degree (X i,2 ), the closer the value is to 1, the more balanced the load; memory usage rate (X i,3 ), in percentage; memory bandwidth (X i,4 ), in GB / s; hard disk read and write speed (X i,5 ), in MB / s; network traffic (X i,6 ), in Mbps; response time (X i,7 ), in ms; temperature (X i,8 ), in °C. At this point, each device has a feature vector X, specifically:

[0043] ;

[0044] where X i represents the feature vector of the i-th device. After obtaining the basic data of the device, preprocessing is required, specifically: first, during the data collection process, data duplicate records may occur, and duplicate items can be removed by detecting the uniqueness of timestamps and device IDs; then, for missing values caused by device failures or sensor problems, interpolation methods at adjacent time points can be used to fill them; next, since different devices and monitoring items may have different dimensions and ranges, the original data cannot be directly used for analysis in the follow-up. For example, the dimensions of network traffic and CPU usage rate are different, so the data of each monitoring item can be standardized using Z-score; finally, using the time series characteristics of the data, time window moving average processing is performed to remove short-term noise.

[0045] S1.2. Calculate the device running fluency and classify to obtain the dataset.

[0046] Furthermore, based on the key indicators of the device, the running state of the device, that is, the fluency F i , is evaluated. The calculation method is as follows:

[0047] ;

[0048] where F idenotes the running smoothness of the $i$-th device; $\exp( )$ represents the exponential function; $\alpha$ represents the adjustment factor. Since the CPU plays a major role during the device operation, $\alpha$ is used to control the influence of the CPU usage rate ($X$ i,1 ) on the running smoothness, which can be set to the empirical value 0.5 in this embodiment; $\log( )$ represents the logarithmic function, which is used to reflect the negative impact of the memory usage rate ($X$ i,3 ) on the device running smoothness; denotes a tiny constant to avoid the denominator being zero, which can be set to the empirical value $1e - 5$ in this embodiment; denotes the hard disk read - write speed ($X$ i,5 ) and the relative relationship with the response time ($X$ i,7 ). When the disk read - write speed is faster and the response time is smaller, the device running smoothness is higher; $\gamma$ represents a tiny constant to avoid the denominator being zero, which can be set to the empirical value $1e - 6$ in this embodiment.

[0049] After obtaining the running smoothness of all devices, the running smoothness can be normalized, that is, using the maximum - minimum normalization method to scale all running smoothness values to the range $[0, 1]$, and the normalized running smoothness is denoted as , that is, the normalized running smoothness of the $i$-th device is , which can also be denoted as the standard running smoothness. Here, classification is performed based on the standard running smoothness of all devices. Set the smoothness threshold and the stuttering threshold . When the standard running smoothness of the device is greater than or equal to the smoothness threshold , it is considered that the device runs smoothly; when the standard running smoothness of the device is less than the smoothness threshold and greater than or equal to the stuttering threshold , it is considered that the device runs relatively smoothly; when the standard running smoothness of the device is less than the stuttering threshold , it is considered that the device runs stutteringly. In this embodiment, the smoothness threshold can be set as , and the stuttering threshold . Exemplary illustration: There is a device with the standard running smoothness being 0.5, which is greater than the stuttering threshold and less than the smoothness threshold , so it can be determined that the device runs relatively smoothly at the current moment. Denote the classification corresponding to the $i$-th device as $Y$ i . Thus, each device has a feature vector and a class label. Then, the feature parameters of all $n$ devices collected at the same moment form a dataset $D(n)$, and there is:

[0050] ;

[0051] Among them, X n,1 represents the CPU usage rate of the nth device; X n,2 represents the load balancing degree of the nth device; X n,8 represents the temperature of the nth device; Y n represents the category label corresponding to the nth device.

[0052] S2. Perform LDA dimensionality reduction processing on the data set based on step S1; obtain the sequence of feature parameters, calculate the concentration degree of each sequence of feature parameters; obtain the synchronous volatility between the sequences of feature parameters and calculate the covariance index to obtain the feature weight matrix of the data set corresponding to each category of devices, and obtain the feature weighted within-class scatter matrix.

[0053] In step S1, a standardized, cleaned data set D(n) containing category labels is obtained, which includes monitoring information of multiple dimensions of various devices. In this embodiment, high-dimensional data will be mapped to a low-dimensional space subsequently for further analysis and processing. The purpose of data dimensionality reduction is to reduce dimensional redundancy and extract the most representative features in the data, especially the features closely related to device status, fault prediction, etc., so as to simplify the analysis model and improve the subsequent processing efficiency. Therefore, the LDA (Linear Discriminant Analysis) algorithm can be used to perform dimensionality reduction processing on the data.

[0054] S2.1. Perform LDA dimensionality reduction processing on the data set based on step S1.

[0055] The LDA algorithm is a supervised learning method, usually used for dimensionality reduction in classification problems. In high-dimensional data, LDA projects the data into a low-dimensional space by finding the projection direction that maximizes the between-class difference. This algorithm is particularly suitable for training data with labels, which can effectively reduce the within-class scatter and increase the between-class scatter at the same time, thus improving the accuracy of data classification. Specifically, LDA achieves dimensionality reduction through the following two matrices, namely: within-class scatter matrix: measuring the differences between samples of the same category; between-class scatter matrix: measuring the differences between different categories.

[0056] In step S1, the data set D(n) is obtained. Based on the category labels of this data set, LDA first calculates the within-class scatter matrix S W , and its calculation formula is:

[0057] ;

[0058] where S W represents the within-class scatter matrix; k represents the total number of categories; represents the category data set; X represents the feature vector of any device in the category data set; represents the The mean vector of the class; T represents the transpose matrix. Then calculate the between-class scatter matrix S B , which is used to measure the differences between different classes, and the formula is:

[0059] ;

[0060] where, S B represents the between-class scatter matrix, k represents the total number of classes; represents the number of samples in the th class; is the mean vector of the samples in the th class; μ is the global mean vector of all class samples; T represents the transpose matrix.

[0061] By calculating the within-class scatter matrix S W and the between-class scatter matrix S B , a new matrix S W -1 S B can be obtained. Solve the eigenvalues and eigenvectors of this matrix, select the eigenvector with the largest eigenvalue, arrange them in size, and select the first t eigenvectors as the reduced-dimensional feature space; project the original data into the feature space composed of the selected t eigenvectors to obtain the reduced-dimensional data.

[0062] After dimensionality reduction by the LDA algorithm, the dimension of the original data is reduced from m to t. In the present invention, the dimension of the original data is m = 8, and the reduced dimension t can be set to the empirical value 2. At this time, the reduced-dimensional data is more concise and more classifiable, facilitating subsequent tasks such as intelligent operation and maintenance analysis.

[0063] S2.2. Obtain the feature parameter sequences and calculate the concentration of each feature parameter sequence.

[0064] The data after LDA dimensionality reduction can be used to detect the abnormal state of the device. By analyzing the distribution of the device in the low-dimensional space, devices deviating from the normal state can be quickly discovered, and timely alarms and processing can be carried out. However, when dealing with multi-dimensional data with high correlations, the LDA algorithm may not be able to effectively capture the discrimination of some key features, especially when the device parameters are highly correlated. This correlation may cause the results after LDA dimensionality reduction to not fully reflect the true differences in the device state, thus affecting the accuracy of subsequent fault prediction and anomaly detection. It is mainly reflected in that when calculating the within-class scatter matrix in the LDA algorithm, due to the collinearity between highly correlated features, the projection direction of the data space is not ideal. Especially in high-dimensional data sets, multiple features may jointly affect the smooth operation of the device, and LDA does not handle the complex relationships between these features well.

[0065] In this embodiment, the operating state of the device is jointly determined by multiple characteristic parameters. There is often a high degree of correlation among these characteristic parameters. Especially in the cases of high device operating load, abnormal ambient temperature, etc., there is often a possibility of device failure. For example, the CPU load and memory usage rate of the device are usually closely related, and the temperature also affects the device state mutually.

[0066] Specifically, in step S1, the dataset D(n) is obtained. Here, the data in the dataset are divided into different category datasets according to different category labels. For the category dataset, it can be denoted as:

[0067] ;

[0068] where represents the dataset of a total of d devices belonging to the category. In this embodiment, since the categories of the devices are set as "smooth", "relatively smooth", and "laggy", so here the maximum value is 3, that is, there are a total of 3 category datasets. In order to overcome the limitations of the LDA algorithm in dealing with highly correlated features, in this embodiment, the calculation method of the within-class scatter matrix S W is adjusted by introducing a weighting factor to strengthen the influence of the characteristic parameters with relatively high correlation of device operation smoothness in the dimensionality reduction process, thereby improving the data representation after dimensionality reduction and enhancing the sensitivity and accuracy of the operation and maintenance management system to device state changes.

[0069] In this embodiment, the device characteristic parameters belonging to the same category should have some common characteristics. For example, in the "smooth" category, the CPU usage rate of all devices should be relatively low and the difference is small, the memory usage rate should also be low and relatively uniform, and the response speed is fast, etc. These characteristics indicate that the parameters themselves have a certain regularity. Therefore, for the dataset of each category in this embodiment, multiple characteristic parameter sequences are extracted. Exemplarily, for the category device dataset , the CPU usage rate sequence of this dataset is extracted, which is , where i represents the i-th device, represents the total number of devices in the category. Correspondingly, there are m characteristic parameter sequences in the dataset , that is:

[0070] To obtain the regularity of the feature parameters themselves, the concentration degree of the feature parameter sequences within each category can be evaluated. For example, in the "smooth" category, the differences in the feature parameter sequences such as CPU usage, memory usage, and response time should be small, so they have a certain similarity.

[0071] For the dataset of the j-th feature parameter sequence of the j class of devices, its concentration degree is denoted as H

[0072] ;

[0073] where H j represents the concentration degree of the j-th feature parameter sequence in the dataset ; Var( ) represents the variance calculation function; represents the dataset j-th feature parameter sequence in j ; μ represents the mean value of the j-th feature parameter sequence in the dataset j When H is smaller, it indicates that the feature parameter sequence is more stable and the differences are smaller, that is, the differences in this feature parameter of all devices in the same category are smaller, and the greater the contribution to the feature vector matrix when calculating the within-class scatter matrix later.

[0074] S2.3. Obtain the synchronous volatility between the feature parameter sequences and calculate the covariance index to obtain the feature weight matrix of the corresponding datasets of each category of devices, and obtain the feature-weighted within-class scatter matrix.

[0075] By calculating the DTW distance between the j-th feature parameter sequence and the p-th feature parameter sequence in the dataset to reflect the similarity between the two feature parameter sequences, so as to evaluate whether there is a certain correlation between the two feature parameter sequences. Here, the DTW distance between the j-th feature parameter sequence and the p-th feature parameter sequence is denoted as Dtw(j, p), and further calculate the synchronous volatility between the two feature parameter sequences. The calculation method is:

[0076] ;

[0077] where R(j, p) represents the synchronous volatility between the j-th feature parameter sequence and the p-th feature parameter sequence; max( ) and min( ) represent the functions of taking the maximum value and the minimum value respectively; represents the j-th feature parameter sequence; represents the p-th feature parameter sequence. The synchronous volatility is used to measure the fluctuation range ratio of two feature parameters in the dataset of the same category. When is closer to When the numerator and denominator are more similar, the larger and closer to 1 R(j,p) is at this time, it indicates that the fluctuation ranges of the two characteristic parameters are similar, which means there is a strong synchronous fluctuation between them; when and have a greater difference, at this time the numerator is much smaller than the denominator, making R(j,p) smaller and closer to 0, which indicates that the difference in the fluctuation ranges of the two characteristic parameters is greater, which means the synchronous fluctuation between them is weaker.

[0078] Through the calculation method of synchronous fluctuation, the similarity measurement between the characteristic sequences is further strengthened. In traditional technologies, the calculation of the correlation between characteristics often ignores the importance of fluctuation consistency in the evaluation of the device operation state. By taking synchronous fluctuation as a new measurement criterion, the present invention can consider the fluctuation consistency of the characteristic sequences during device state classification, thereby more accurately reflecting the operation status of the device and effectively avoiding the misjudgment problems that may occur in traditional methods.

[0079] Finally, calculate the covariation index between each pair of characteristic parameter sequences, and the specific calculation method is:

[0080] ;

[0081] where U(j,p) represents the covariation index between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence; R(j,p) represents the synchronous fluctuation between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence; Dtw(j,p) represents the DTW distance between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence; σ represents a small constant to avoid the denominator being 0, and in this embodiment, an empirical value of 1e-4 can be taken.

[0082] When R(j,p) is larger and Dtw(j,p) is smaller, it indicates that the fluctuation ranges between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence are similar and the sequence similarity is larger, then the covariation index U(j,p) between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence is larger, and at this time the covariation between these two characteristic parameter sequences in the same category dataset is stronger; while when R(j,p) is smaller and Dtw(j,p) is larger, it indicates that the difference in the fluctuation ranges between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence is larger and the sequence similarity is smaller, then the covariation index U(j,p) between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence is smaller, and at this time the covariation between these two characteristic parameter sequences in the same category dataset is weaker.

[0083] By introducing a calculation method for the co-variability index, the mutual relationship between characteristic parameter sequences can be measured more precisely. Compared with the traditional covariance method, the present invention calculates the co-variability between features by combining a comprehensive index of synchronous volatility and DTW distance, which can effectively handle the multi-dimensional dependence relationship of the device in a complex operating environment, thereby improving the dimensionality reduction effect of the LDA algorithm in high-dimensional data and enhancing the accuracy of device state analysis.

[0084] Thus, the concentration degree H of the j-th characteristic parameter sequence is obtained. j And the co-variability index U(j, p) between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence. Finally, calculate at the class data set the weight of the j-th characteristic parameter, and the calculation method is:

[0085] ;

[0086] where represents the weight of the j-th characteristic parameter in the data set ; represents the co-variability index between the remaining characteristic parameter sequences and the j-th characteristic parameter sequence in the class data set except the j-th characteristic parameter sequence; H j represents the concentration degree of the j-th characteristic parameter sequence in the data set ; β represents a small constant to avoid the denominator being zero. In this embodiment, the empirical value 1e-3 can be taken. The same as the calculation method of the weight of the j-th characteristic parameter, obtain the weights of the remaining characteristic parameters in the data set , and the characteristic parameter weight matrix can be obtained, and there is:

[0087] ;

[0088] where represents the characteristic parameter weight matrix of the class data set ; represents a total of m characteristic parameter weights. In this embodiment, there is . After obtaining the characteristic parameter weight matrix , add it to the calculation formula of the within-class scatter matrix to weight the scatter contribution of each feature, thereby adjusting the influence of different features. The weighted within-class scatter matrix is:

[0089] ;

[0090] where, S * W represents the feature-weighted within-class scatter matrix; k represents the total number of categories; denotes the class dataset; denotes the class dataset feature parameter weight matrix; X denotes the feature vector of any device in the denotes the class mean vector; T denotes the transpose matrix. The improved within-class scatter matrix S * W is compared with the between-class scatter matrix S B to optimize the projection direction of the data. By solving the improved scatter matrix S * W -1 S B the feature vector is obtained, and the feature vector with the largest eigenvalue is selected as the basis vector for dimensionality reduction, and finally the data is projected into the low-dimensional space.

[0091] Through the calculation of the feature-weighted within-class scatter matrix, the process of LDA dimensionality reduction is further refined. Compared with the traditional LDA algorithm that only relies on the within-class scatter matrix, this method considers the weighted influence of device features, thus enhancing the classification effect of the dimensionality reduction process. This method enables the features of the device to be more accurately represented according to the class differences, improving the recognition ability of the intelligent operation and maintenance system for the device status and the accuracy of anomaly warning.

[0092] S3. Complete the information-based intelligent operation and maintenance of the device based on the dimensionality-reduced data.

[0093] After the data dimensionality reduction is completed in step S2, a dataset containing the dimensionality-reduced feature space is obtained. These data are processed by the LDA (Linear Discriminant Analysis) algorithm, removing redundant and irrelevant features and retaining the key features for classifying the device operating status. At this time, the dimension of the data has been reduced from 8 dimensions to 2 dimensions. The dimensionality-reduced data can be effectively used for subsequent intelligent operation and maintenance tasks, especially in aspects such as fault prediction, anomaly detection, and visualization display.

[0094] In one embodiment, by further analyzing and modeling this data, it is possible to achieve the prediction of equipment failures. Exemplary illustration: Use the dataset after dimensionality reduction as the training set, and apply common classification models (such as Support Vector Machine (SVM), Random Forest, or Neural Network) to train the prediction model of the equipment's operating state. Since the data dimension has been reduced to a lower level, the training speed and effect of the model will be improved. And the model will be trained according to category labels (such as "smooth", "relatively smooth", and "laggy") so that the model can identify different operating states of the equipment. During daily operation and maintenance, new equipment data can be predicted through the trained model. The model will output whether the equipment is in a normal working state or predict whether the equipment is about to fail based on the feature data after dimensionality reduction. For example, when the distribution of equipment features in the reduced-dimensional space changes significantly, the model can prompt that the equipment may fail and perform maintenance in advance.

[0095] In the second embodiment, the data after dimensionality reduction can also be used for anomaly detection to help identify abnormal equipment states in real time and take corresponding emergency measures. Exemplary illustration: Clustering algorithms (such as K-means clustering or DBSCAN) can be used to cluster the equipment feature data after dimensionality reduction to form multiple operating state clusters of the equipment. When the Euclidean distance between the parameters of the equipment corresponding to the reduced dimensions and the parameters of other equipment in the cluster is greater than the set threshold empirical value, it indicates that the equipment may be in an abnormal state. Through the clustering results, the system can quickly identify and isolate the equipment with significant differences in features from other equipment, further prompt whether there are potential faults in the equipment, and notify the corresponding operation and maintenance personnel to confirm the equipment operating state.

[0096] In the third embodiment, the data after dimensionality reduction makes the operating state of the equipment more visual. By mapping the data into a 2D space, it can help operation and maintenance personnel intuitively understand the operating conditions and fault warnings of the equipment. Exemplary illustration: By mapping the data after dimensionality reduction into a 2D space, visualization tools (such as Matplotlib or Tableau) can be used to generate scatter plots or heat maps of the equipment's running smoothness. These graphs can help operation and maintenance personnel quickly identify the state of the equipment, check which equipment is within the normal range, and which equipment is about to fail or has already failed.

[0097] The present invention classifies and reduces the dimension of data by combining the running smoothness of the device, significantly improving the accuracy and efficiency of the information-based intelligent operation and maintenance method. Compared with the prior art, traditional operation and maintenance methods often rely on single features or simple statistical indicators to evaluate the running state of the device, while the present invention calculates the running smoothness of the device by integrating the running parameters of multiple devices and divides the devices into different categories based on the smoothness. Especially in the LDA dimension reduction process, the introduction of the feature-weighted within-class scatter matrix further enhances the sensitivity of the algorithm to the relationship between features, making the dimension-reduced data more concise and having a higher class discrimination degree. By using the clustering algorithm to further identify whether there is an abnormality in the device, it can more efficiently and accurately warn of faults. This method breaks through the limitations of traditional operation and maintenance methods, can process complex device data, and improves the accuracy and response speed of device state recognition in the dimension reduction and clustering processes, and is particularly suitable for the intelligent operation and maintenance environment where the device state changes frequently and complexly.

[0098] Embodiment of the information-based intelligent operation and maintenance system:

[0099] On the other hand, the present invention also provides an information-based intelligent operation and maintenance system. As Figure 2 shown, the information-based intelligent operation and maintenance system includes a processor and a memory, and the memory stores computer program instructions, which, when executed by the processor, implement an information-based intelligent operation and maintenance method according to the first aspect of the present invention.

[0100] The information-based intelligent operation and maintenance system further includes other components well-known to those skilled in the art such as a communication interface, and its settings and functions are known in the art, so they will not be described in detail here.

[0101] In the present invention, the foregoing memory may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, the computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory (RRAM), a dynamic random access memory (DRAM), a static random access memory (SRAM), an enhanced dynamic random access memory (EDRAM), a high-bandwidth memory (HBM), a hybrid memory cube (HMC), and so on, or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that can be stored or otherwise held by such a computer-readable medium.

[0102] In the description of this specification, the meanings of "a plurality of" and "several" are at least two, such as two, three, or more, etc., unless otherwise specifically defined.

[0103] Although this specification has shown and described multiple embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and scope of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in the practice of the present invention.

Claims

1. An information-based intelligent operation and maintenance method, characterized in that, The method includes: Collecting the operation parameters of multiple devices and calculating the operation smoothness, classifying each device based on the operation smoothness to obtain the category of each device; the operation parameters of all devices form a data set, and the data set is dimensionally reduced using the LDA algorithm based on the category to obtain the dimensionally reduced data; clustering the data to obtain a clustering result, and judging whether there is an abnormality in the device based on the clustering result; The LDA algorithm further includes processing the within-class scatter matrix to obtain a feature-weighted within-class scatter matrix, and the feature-weighted within-class scatter matrix is the product of the feature weight matrix and each feature vector in the within-class scatter matrix; the calculation method of the feature weight matrix is specifically: Dividing the data set into multiple classified data sets according to the device category, sorting the same operation parameter of all devices in the classified data set according to the device number to obtain a corresponding feature parameter sequence; calculating the ratio of the variance to the mean of the feature parameter sequence to obtain the concentration; Calculating the absolute value of the difference between the maximum value and the minimum value in each feature parameter sequence; for any two feature parameter sequences, taking the ratio of twice the corresponding absolute value minimum to the sum of the corresponding absolute values as the synchronous volatility; and calculating the ratio of the synchronous volatility to the DTW distance between the two feature parameter sequences to obtain the co-variability index; taking the ratio of the sum of the co-variability indexes of each feature parameter sequence and the other feature parameter sequences to the concentration of each feature parameter sequence as the weight of the corresponding feature, and the weights of all features form a feature weight matrix.

2. The information-based intelligent operation and maintenance method according to claim 1, wherein The calculation method of the feature-weighted within-class scatter matrix is specifically: ； Among them, S * W represents the feature weighted within-class scatter matrix; k represents the total number of classes; represents the class data set; represents the feature parameter weight matrix of the class data set; X represents the feature vector of any device in the class data set; represents the class mean vector; T represents the transpose matrix.

3. The information-based intelligent operation and maintenance method according to claim 1, characterized in that The calculation method of the weight is specifically: ； wherein represents the weight of the j-th characteristic parameter in the -th class of data sets; represents the co-variability index of the remaining characteristic parameter sequences and the j-th characteristic parameter sequence in the -th class of data sets except for the j-th characteristic parameter sequence; H j represents the concentration of the j-th characteristic parameter sequence in the -th class of data sets; β represents a small constant.

4. The information-based intelligent operation and maintenance method according to claim 3, wherein The calculation method of the co-variability index is specifically: ； Where U(j,p) represents the co-variability index between the jth feature parameter sequence and the pth feature parameter sequence; R(j,p) represents the synchronous volatility between the jth feature parameter sequence and the pth feature parameter sequence; Dtw(j,p) represents the DTW distance between the jth feature parameter sequence and the pth feature parameter sequence; σ represents a tiny constant.

5. An information-based intelligent operation and maintenance method according to claim 4, wherein, The calculation method of the synchronous volatility is specifically: ； Among them, R(j, p) represents the synchronous volatility between the j-th characteristic parameter sequence and the p-th characteristic parameter sequence; max( ) and min( ) represent the functions of taking the maximum value and the minimum value respectively; represents the j-th characteristic parameter sequence; represents the p-th characteristic parameter sequence.

6. The information-based intelligent operation and maintenance method according to claim 1, characterized in that The calculation method of the operation smoothness is specifically: ； Among which F i represents the running smoothness of the i-th device; exp( ) represents the exponential function; α represents the adjustment factor; (X i,1 ) represents the CPU usage rate; log( ) represents the logarithmic function; (X i,3 ) represents the memory usage rate; represents a tiny constant; represents the relative relationship between the hard disk read / write speed (X i,5 ) and the response time (X i,7 ); γ represents a tiny constant.

7. An information-based intelligent operation and maintenance method according to claim 6, characterized in that, Classifying each device based on the operation smoothness, including: Responding to the operation smoothness of the device being greater than or equal to the smoothness threshold, classifying the device as operating smoothly; Responding to the operation smoothness of the device being greater than or equal to the lag threshold and less than the smoothness threshold, classifying the device as operating relatively smoothly; Responding to the operation smoothness of the device being less than the lag threshold, classifying the device as operating laggy.

8. An information-based intelligent operation and maintenance method according to claim 6, characterized in that, It also includes preprocessing the operation parameters of the collected multiple devices, specifically: Removing duplicate operation parameters by detecting the uniqueness of the time stamp and the device ID; Filling in the missing values of the operation parameters using the interpolation method of adjacent time points; Standardizing the operation parameters using the Z-score algorithm; Removing short-term noise from the operation parameters by time window sliding average processing.

9. An information-based intelligent operation and maintenance method according to claim 1, characterized in that Clustering the data to obtain a clustering result, and judging whether there is an abnormality in the device based on the clustering result, including: Using the K-Means algorithm to cluster the data to obtain multiple clustering clusters; Calculate the Euclidean distance between each data and the data of the corresponding clustering cluster center; In response to the Euclidean distance being greater than the set distance threshold, determine that the corresponding device is abnormal and notify the operation and maintenance personnel to confirm the status.

10. An information-based intelligent operation and maintenance system, characterized in that, It includes a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, the information-based intelligent operation and maintenance method described in any one of claims 1 to 9 is implemented.

Images