Equipment fault prediction method and device, storage medium and computer program product

Abstract

The invention discloses an equipment fault prediction method and device, a storage medium and a computer program product. The method comprises the following steps: acquiring operation state time sequence data of equipment; according to first similarity and second similarity corresponding to the operation state time sequence data, design parameters of a convolution kernel are determined, the first similarity is used for representing the correlation degree between the index values of all the dimensions, and the second similarity is used for representing the correlation degree between the index values of all the dimensions and the abnormal point sequence; constructing a target convolution kernel according to the design parameters, and performing feature extraction on the running state time sequence data by adopting the target convolution kernel to obtain target features; and determining the fault risk probability of the equipment according to the target characteristics. According to the method and the device, the technical problem of poor equipment fault prediction effect caused by incomplete feature capture when time series data is analyzed due to single-size convolution kernel design in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of data analysis technology, and in particular to a method, device, storage medium and computer program product for predicting equipment failure. Background Art

[0002] Convolutional neural network is a powerful deep learning model that has not only achieved great success in the field of image processing, but also plays an important role in time series prediction tasks. For example, in equipment failure prediction, by obtaining multi-dimensional indicator data with time stamps related to the equipment, this type of data is usually large in volume, with chaotic time series changes and difficult to summarize manually. This scenario is suitable for related processing through convolutional neural network technology, and the extraction of related features by convolution kernel is the core key.

[0003] In the related art, multiple columns of features are processed simultaneously through multiple convolution kernels of the same size to extract data features. However, the single-size convolution kernel design in the related art makes it difficult to obtain features of different time windows, and the feature capture is incomplete when analyzing time series data, resulting in problems such as poor equipment failure prediction effect.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a device fault prediction method, apparatus, storage medium and computer program product to at least solve the technical problem of poor device fault prediction effect caused by incomplete feature capture when analyzing time series data due to the single-size convolution kernel design in the related technology.

[0006] According to one aspect of an embodiment of the present application, a method for predicting equipment failure is provided, including: obtaining equipment operation status time series data, wherein the operation status time series data includes index values ​​of multiple dimensions corresponding to the equipment at multiple data collection time points; determining design parameters of a convolution kernel based on a first similarity and a second similarity corresponding to the operation status time series data, wherein the first similarity is used to characterize the degree of correlation between the index values ​​of each dimension, and the second similarity is used to characterize the degree of correlation between the index values ​​of each dimension and an abnormal point sequence, and the abnormal point sequence is used to indicate a data collection time point at which an abnormal index value in the operation status time series data exists; constructing a target convolution kernel according to the design parameters, and using the target convolution kernel to perform feature extraction on the operation status time series data to obtain a target feature; determining the equipment failure risk probability based on the target feature, wherein the failure risk probability is used to characterize the probability of the equipment failing in the future.

[0007] Optionally, the running status time series data contains multiple indicator data columns, each indicator data column is an indicator value arranged in time series order under one dimension, and the design parameters of the convolution kernel include: convolution kernel width and convolution kernel length; according to the first similarity and the second similarity corresponding to the running status time series data, determining the design parameters of the convolution kernel includes: determining the first similarity corresponding to each indicator data column in the running status time series data, and determining the convolution kernel width based on the first similarity; determining the second similarity between each indicator data column in the running status time series data and the abnormal point sequence, and determining the convolution kernel length based on the second similarity.

[0008] Optionally, based on the first similarity, determining the convolution kernel width includes: calculating the Pearson correlation coefficient between each pair of indicator data columns, and determining the Pearson correlation coefficient as the first similarity; determining the number of neighbors corresponding to each indicator data column, wherein the number of neighbors is the number of other indicator data columns whose first similarity with the indicator data column is greater than or equal to a preset similarity threshold; determining the minimum number of neighbors among the numbers of neighbors corresponding to each indicator data column, and when the value of the minimum number of neighbors is greater than 1, determining the value of the minimum number of neighbors as the convolution kernel width.

[0009] Optionally, the method also includes: selecting an indicator data column with the largest number of neighbors as the first column in the running status time series data; selecting a target number of indicator data columns from other indicator data columns that have not been selected in the running status time series data in order from large to small for arrangement after the first column, wherein the target number is the value of the convolution kernel width minus 1; determining the last indicator data column that has been selected and sorted as the last column, and continuing to select a target number of indicator data columns from other indicator data columns that have not been selected in order from large to small for arrangement according to the first similarity between the first similarity and the last column, until all the indicator data columns in the running status time series data have been selected and arranged.

[0010] Optionally, determining the convolution kernel length based on the second similarity includes: obtaining an outlier point sequence, wherein the outlier point sequence includes label values ​​corresponding to each data collection time point arranged in chronological order, wherein when the label value is a first numerical value, it represents that there is an anomaly in the indicator value corresponding to the data collection time point, and when the label value is a second numerical value, it represents that there is no anomaly in the indicator value corresponding to the data collection time point; calculating the Kendall correlation coefficient between the outlier point sequence and each indicator data column, and determining the Kendall correlation coefficient as the second similarity; determining the indicator data column with the largest absolute value of the second similarity in the operating status time series data as the benchmark data column; determining the period length corresponding to the benchmark data column, and determining the convolution kernel length based on the period length.

[0011] Optionally, the convolution kernel length includes: a baseline convolution kernel length, a first convolution kernel length, and a second convolution kernel length; determining the convolution kernel length based on the cycle length includes: performing a Fourier transform on the baseline data column to obtain frequency domain data corresponding to the baseline data column, wherein the Fourier transform is used to convert the baseline data column from the time domain to the frequency domain; determining the main frequency with the largest power in the frequency domain data, and determining the cycle length corresponding to the baseline data column based on the main frequency; determining the cycle length as the baseline convolution kernel length, and expanding the first convolution kernel length based on a first proportional coefficient and a second proportional coefficient to obtain a first convolution kernel length and a second convolution kernel length, wherein the first proportional coefficient is less than 1 and the second proportional coefficient is greater than 1.

[0012] Optionally, a target convolution kernel is used to extract features from the running status timing data to obtain target features, including: constructing a first convolution kernel based on the convolution kernel width and the first convolution kernel length, constructing a second convolution kernel based on the convolution kernel width and the second convolution kernel length, and constructing a third convolution kernel based on the convolution kernel width and the reference convolution kernel length; using the first convolution kernel, the second convolution kernel, and the third convolution kernel to respectively extract features from the running status timing data to obtain a first feature, a second feature, and a third feature; determining the target feature based on the first feature, the second feature, and the third feature.

[0013] According to another aspect of an embodiment of the present application, there is also provided a device fault prediction apparatus, including: a time series data acquisition module, used to acquire the operating status time series data of the device, wherein the operating status time series data contains index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; a convolution kernel design module, used to determine the design parameters of the convolution kernel based on the first similarity and the second similarity corresponding to the operating status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index values ​​of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the operating status time series data is abnormal; a feature extraction module, used to construct a target convolution kernel according to the design parameters, and use the target convolution kernel to perform feature extraction on the operating status time series data to obtain a target feature; a fault prediction module, used to determine the failure risk probability of the device based on the target feature, wherein the failure risk probability is used to characterize the probability of the device failing in the future.

[0014] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, the non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the device fault prediction method by running the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, which implements the steps of the device failure prediction method when the computer program is executed by a processor.

[0016] In an embodiment of the present application, the operating status time series data of the device is obtained, wherein the operating status time series data contains index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; the design parameters of the convolution kernel are determined based on the first similarity and the second similarity corresponding to the operating status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index values ​​of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the operating status time series data is abnormal; a target convolution kernel is constructed according to the design parameters, and the target convolution kernel is used to extract features from the operating status time series data to obtain target features; based on the target features, the failure risk probability of the device is determined, wherein the failure risk probability is used to characterize the probability of the device failing in the future. By adaptively adjusting the convolution kernel based on the time series data, the purpose of improving the feature extraction effect of the convolution neural network in processing the time series data prediction task is achieved, and the prediction accuracy of the equipment failure prediction is improved, thereby solving the technical problem of poor equipment failure prediction effect caused by incomplete feature capture when analyzing the time series data due to the single-size convolution kernel design in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 It is a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for predicting equipment failures according to an embodiment of the present application;

[0019] Figure 2 It is a schematic diagram of a method flow for predicting equipment failures provided in accordance with an embodiment of the present application;

[0020] Figure 3 is a schematic diagram of operating status time series data provided according to an embodiment of the present application;

[0021] Figure 4 It is a schematic diagram of a process of performing feature extraction using a convolution kernel corresponding to a convolution kernel design solution in the related art provided in an embodiment of the present application;

[0022] Figure 5It is a schematic diagram of a correlation coefficient matrix between two indicator data columns provided according to an embodiment of the present application;

[0023] Figure 6 is a schematic diagram of a convolution processing process provided according to an embodiment of the present application;

[0024] Figure 7 It is a structural schematic diagram of an equipment failure prediction device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] In time series prediction task scenarios such as equipment failure prediction, traffic prediction, and transaction data inspection, it is suitable to use convolutional neural network technology for related processing. The convolution kernel is the core component of the convolutional neural network and the basic unit for performing convolution operations. It plays a decisive role in the performance and feature extraction capabilities of the network. The convolution kernel slides on the input data and calculates the weighted sum of the local area to extract the local features of the input data. The design of the convolution kernel includes factors such as size and number, which will affect the performance and training effect of the network. Generally speaking, the size of the convolution kernel is usually an odd number, such as 3x3, 5x5, 7x7, etc. Smaller convolution kernels can extract more local features, while larger convolution kernels can extract more global features.

[0028] The convolution kernel design method in the related art mainly processes multiple columns of features simultaneously through multiple convolution kernels of the same size to extract data features. However, when processing time series data through convolutional neural networks, due to the periodicity and volatility of time series data, the single-size convolution kernel design in the related art is difficult to obtain features of different time windows. In addition, the convolution kernel design in the related art does not consider the correlation of dimensions (columns). After weighted and calculated, the features between different dimensions (columns) will be covered.

[0029] Therefore, when predicting time series data, the use of traditional convolution kernels in convolutional neural networks in related technologies will result in incomplete acquisition of features in different time windows, and some features will be covered up due to weighting and calculation.

[0030] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.

[0031] According to an embodiment of the present application, a method embodiment of equipment fault prediction is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or electronic device) for implementing a method for predicting equipment failure. Figure 1 As shown, the computer terminal 10 (or electronic device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0033] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or electronic device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0034] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the device fault prediction method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned device fault prediction method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0035] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0036] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or electronic device).

[0037] Under the above operating environment, the embodiment of the present application provides a method for predicting equipment failures. Figure 2 is a schematic diagram of a method flow for predicting equipment failures provided in an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0038] Step S202, obtaining the time series data of the running status of the device, wherein the time series data of the running status includes the index values ​​of multiple dimensions corresponding to the device at multiple data collection time points;

[0039] Step S204, determining the design parameters of the convolution kernel according to the first similarity and the second similarity corresponding to the running status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index value of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the running status time series data is abnormal;

[0040] Step S206, constructing a target convolution kernel according to the design parameters, and using the target convolution kernel to extract features from the operating state time series data to obtain target features;

[0041] Step S208, determining the failure risk probability of the equipment based on the target characteristics, wherein the failure risk probability is used to characterize the probability of the equipment failing in the future.

[0042] Through the above steps, by adaptively adjusting the convolution kernel according to the time series data, the purpose of improving the feature extraction effect of the convolutional neural network in processing the time series data prediction task is achieved, and the prediction accuracy of the equipment fault prediction is improved, thereby solving the technical problem of poor equipment fault prediction effect caused by incomplete feature capture when analyzing time series data due to the single-size convolution kernel design in the related technology.

[0043] The following further introduces the equipment failure prediction method in steps S202 to S208 of the embodiment of the present application.

[0044] First, the operating status time series data is obtained. For example, in this embodiment, convolution processing is performed on the same set of time series data (25,000 items) to compare the accuracy of the traditional method in the related art and the method of the present invention.

[0045] The following is Figure 3 The data sample shown is taken as an example to illustrate the processing flow in this application, where each row of data (t443, t444...t454) is the data collection time point, each column of data (f_0, f_1, f_2, f_3) is the indicator value of 4 different dimensions corresponding to the device (f_0, f_1, f_2, f_3 are 4 indicator data columns), and column Y is the abnormal point sequence.

[0046] For the traditional method in the related technology, the dimension of the time series data is usually selected as the convolution kernel width. Figure 3The data shown (f_0, f_1, f_2, f_3), the convolution kernel width used in the related art is set to 4, while the length and width of the traditional convolution kernel are consistent, so for Figure 3 In the 12×4 data domain, the related technology obtains a 9*1 convolution result by convolution calculation with a 4×4 convolution kernel of 1 step length, such as Figure 4 shown.

[0047] The specific process of convolution kernel design and feature extraction of this application is as follows.

[0048] After obtaining the running status timing data, the convolution kernel can be adaptively designed based on the data. The specific steps are as follows.

[0049] In some embodiments of the present application, the operating status time series data includes multiple indicator data columns, each indicator data column is an indicator value arranged in a time series order under one dimension, and the design parameters of the convolution kernel include the following steps: convolution kernel width, convolution kernel length; according to the first similarity and the second similarity corresponding to the operating status time series data, determining the design parameters of the convolution kernel includes: determining the first similarity corresponding to each indicator data column in the operating status time series data, and determining the convolution kernel width based on the first similarity; determining the second similarity between each indicator data column in the operating status time series data and the abnormal point sequence, and determining the convolution kernel length based on the second similarity.

[0050] Specifically, in this embodiment, the minimum neighbor value can be selected as the convolution kernel width through the feature similarity (i.e., the first similarity) and the neighbor value between each indicator data column; and, by calculating the correlation between the indicator data column and the labeled outlier point sequence (i.e., the second similarity), the period of the calculated indicator data column with the highest correlation is obtained to determine the convolution kernel length, which are described in detail below.

[0051] Among them, the specific steps for determining the convolution kernel width are as follows.

[0052] In some embodiments of the present application, determining the convolution kernel width based on the first similarity includes the following steps: calculating the Pearson correlation coefficient between each pair of indicator data columns, and determining the Pearson correlation coefficient as the first similarity; determining the number of neighbors corresponding to each indicator data column, wherein the number of neighbors is the number of other indicator data columns whose first similarity with the indicator data column is greater than or equal to a preset similarity threshold; determining the minimum number of neighbors among the numbers of neighbors corresponding to each indicator data column, and when the value of the minimum number of neighbors is greater than 1, determining the value of the minimum number of neighbors as the convolution kernel width.

[0053] Specifically, in this embodiment, the feature approximation (first similarity) r between two dimensions (indicator data columns) can be calculated by the Pearson correlation coefficient, where r is used to measure the correlation (linear correlation) between two variables X and Y, and its value is between -1 and 1. When the absolute value of the correlation coefficient is close to 1, it indicates that there is a strong linear relationship between the two variables (indicator data columns); when the absolute value of the correlation coefficient is close to 0, it indicates that there is almost no linear relationship between the two variables (indicator data columns). The specific calculation formula is shown in the following formula:

[0054]

[0055] Among them, Xi and Yi are the i-th observation values ​​of variables X and Y respectively, and are the means (averages) of variables X and Y, respectively. The numerator in the formula is the sum of the products of the deviations of X and Y, representing the covariance of the deviations of the two variables. The denominator is the square root of the sum of the squares of the deviations of X and Y, representing the standard deviations of the two variables, respectively. In this embodiment, variables X and Y are any two indicator data columns in the operating status time series data.

[0056] For example, in Figure 3 After the first similarity calculation of the running status time series data shown in FIG. 1 is completed, the correlation coefficient matrix between the two correlations of the indicator data columns can be obtained, such as Figure 5 shown.

[0057] Then, by comparing the first similarity between each indicator data column and other indicator data columns and the preset similarity threshold, the number of neighbors k corresponding to each indicator data column is determined. Based on statistical principles, when the Pearson correlation coefficient (first similarity) reaches 0.6, it indicates that there is a moderate positive correlation between the variables, which can reflect the more significant correlation between the data. Therefore, in this embodiment, the preset similarity threshold is set to 0.6. That is, if r>0.6 between the current indicator data column, the indicator data column is the neighbor of the current indicator data column, thereby obtaining the number of neighbors k for each indicator data column. For example, k=1 for f_0, k=1 for f_1, k=0 for f_2, and k=0 for f_3.

[0058] Finally, the minimum value kmin of the number of nearest neighbors k can be selected as the convolution kernel width f; and when kmin<=1, the default value, such as f=2, can be used, and in other cases, f=kmin. For example, since the current set of nearest neighbors is [1,1,0,0], kmin=0, so the final convolution kernel width is the preset default value, that is, f=2.

[0059] In addition, the indicator data columns in the running status time series data may be reordered according to the first similarity and the number of nearest neighbors. The specific steps are as follows.

[0060] In some embodiments of the present application, the method also includes the following steps: selecting an indicator data column with the largest number of neighbors as the first column in the running status time series data; selecting a target number of indicator data columns from other indicator data columns that have not been selected in the running status time series data in order from large to small for arrangement after the first column, wherein the target number is the value of the convolution kernel width minus 1; determining the last indicator data column that has been selected and sorted as the last column, and continuing to select a target number of indicator data columns from other indicator data columns that have not been selected in order from large to small for arrangement, until all the indicator data columns in the running status time series data have been selected and arranged.

[0061] Specifically, first select the indicator data column where the maximum value kmax of the number of nearest neighbors k is located as the first column. If there are 2 or more kmax, then any indicator data column where kmax is located can be selected as the first column; then select the target number (f-1) columns with the highest correlation with the first column, and sort them from large to small according to the first similarity r after the first column; then, and so on, select the target number (f-1) indicator data columns with the highest correlation with the current last indicator data column column by column, and sort the unselected columns after this column according to the first similarity r from large to small until all columns are sorted.

[0062] For example, currently the number of nearest neighbors of f_0 and f_1 is both 1, which is the maximum value, so any column is selected as the first column. In this embodiment, column f_0 is selected as the first column; because f_0 is the first column and the convolution width is 2, an indicator data column with the largest first similarity with f_0 is selected. In this embodiment, f_1 is the nearest neighbor of f_0, so the first step order is f_0, f_1; similarly, except for f_0, f_2 is the nearest neighbor of f_1, and the current order is f_0, f_1, f_2; finally only f_3 is left, which is merged into the feature matrix, and the final order is f_0, f_1, f_2, f_3.

[0063] The following is an introduction to the process of determining the length of the convolution kernel. The specific steps are as follows.

[0064] In some embodiments of the present application, determining the convolution kernel length based on the second similarity includes the following steps: obtaining an outlier point sequence, wherein the outlier point sequence includes label values ​​corresponding to each data acquisition time point arranged in chronological order, wherein when the label value is a first numerical value, it represents that there is an abnormality in the indicator value corresponding to the data acquisition time point, and when the label value is a second numerical value, it represents that there is no abnormality in the indicator value corresponding to the data acquisition time point; calculating the Kendall correlation coefficient between the outlier point sequence and each indicator data column, and determining the Kendall correlation coefficient as the second similarity; determining the indicator data column with the largest absolute value of the second similarity in the operating status time series data as the benchmark data column; determining the period length corresponding to the benchmark data column, and determining the convolution kernel length based on the period length.

[0065] Specifically, first obtain the abnormal point sequence corresponding to the running status time series data, which can be obtained by marking the abnormal points of the data. In this embodiment, the abnormal point sequence is Figure 3 The Y column shown in , where "0" indicates that there is no abnormality in the indicator value corresponding to the data collection time point (row) in the running status time series data, and "1" indicates that there is an abnormality in the indicator value corresponding to the data collection time point (row) in the running status time series data.

[0066] Afterwards, Kendall correlation analysis is used to calculate the correlation (second similarity) R between the dimension (index data column) and the outlier sequence, as shown in the following formula:

[0067]

[0068] Among them, τ is the Kendall correlation coefficient (i.e., the second similarity), n is the number of data collection time points, xi and yi are the values ​​of the i-th data point in the two variables (indicator data series and anomaly point series), respectively, and \sgn()\sgn() is a sign function that returns 1 if its argument is greater than 0, -1 if it is less than 0, and 0 if it is equal to 0.

[0069] For each pair of data points (xi,yi) and (xj,yj), calculate their rank differences in the two variables xi-xj and yi-yj; calculate the product of the sign functions of these two rank differences \sgn(xi-xj)\sgn(yi-yj); add all such products and then multiply by The second similarity R can be obtained.

[0070] For example, the second similarity corresponding to f_0 is -0.14564, the second similarity corresponding to f_1 is -0.04193, the second similarity corresponding to f_2 is 0.021296, and the second similarity corresponding to f_3 is 0.395319.

[0071] Then, the period of the column where the maximum absolute value Rmax of R is located (ie, the above-mentioned reference data column) is selected as the standard period l. Specifically, the Fourier transform method can be used to calculate and determine the period length l corresponding to the reference data column. The specific steps are as follows.

[0072] In some embodiments of the present application, the convolution kernel length includes: a baseline convolution kernel length, a first convolution kernel length, and a second convolution kernel length; determining the convolution kernel length based on the cycle length includes: performing a Fourier transform on the baseline data column to obtain frequency domain data corresponding to the baseline data column, wherein the Fourier transform is used to convert the baseline data column from the time domain to the frequency domain; determining the main frequency with the largest power in the frequency domain data, and determining the cycle length corresponding to the baseline data column based on the main frequency; determining the cycle length as the baseline convolution kernel length, and expanding the first convolution kernel length based on a first proportional coefficient and a second proportional coefficient to obtain a first convolution kernel length and a second convolution kernel length, wherein the first proportional coefficient is less than 1 and the second proportional coefficient is greater than 1.

[0073] For example, according to the above calculation results, f_3 is the column where Rmax is located, that is, f_3 is determined as the reference data column; the Fourier transform method is used to calculate the period length value corresponding to f_3, and the calculation result is 4.

[0074] Then the cycle length is determined as the benchmark convolution kernel length, that is, 4 is determined as the benchmark convolution kernel length, and half cycle, full cycle, and 1.5 cycle are selected as the convolution kernel length, and the convolution kernel lengths are 0.5l, l, and 1.5l respectively, that is, the half cycle, full cycle, and 1.5 cycle values ​​are 0.5*4, 1*4, and 1.5*4 respectively, so the first convolution kernel length, the benchmark convolution kernel length, and the second convolution kernel length are 2, 4, and 6 respectively.

[0075] In addition, if there is no period in the reference data sequence corresponding to Rmax, the preset default convolution kernel length is used, for example, the first convolution kernel length is 3, the reference convolution kernel length is 5, and the second convolution kernel length is 7.

[0076] After determining the design parameters corresponding to the convolution kernel, you can construct the convolution kernel and perform convolution processing on the data. The specific steps are as follows.

[0077] In some embodiments of the present application, a target convolution kernel is used to extract features from operating status timing data, and obtaining target features includes the following steps: constructing a first convolution kernel based on a convolution kernel width and a first convolution kernel length, and constructing a second convolution kernel based on a convolution kernel width and a second convolution kernel length, and constructing a third convolution kernel based on a convolution kernel width and a reference convolution kernel length; using the first convolution kernel, the second convolution kernel, and the third convolution kernel to extract features from the operating status timing data, respectively, to obtain a first feature, a second feature, and a third feature; determining the target feature based on the first feature, the second feature, and the third feature.

[0078] Specifically, the first convolution kernel, the second convolution kernel, and the third convolution kernel are constructed respectively, and the corresponding convolution kernel sizes are 0.5l×f, l×f, and 1.5l×f respectively. In this example, f=2, l=4, so the convolution kernel sizes are 2*2, 4*2, and 6*2 respectively. Afterwards, the first convolution kernel, the second convolution kernel, and the third convolution kernel are used to extract features from the running status time series data respectively, and the processing results are assembled to obtain the final convolution result (target feature). The process is as follows: Figure 6 shown.

[0079] Finally, the failure risk probability of the equipment is determined based on the extracted target features. After testing, 25,000 time series data are used for modeling. After applying the convolution kernel designed in this application scheme, the accuracy of fault risk prediction is improved by 5.33% compared with the traditional method in the related technology, and the improvement effect is obvious.

[0080] The convolution kernel design method of the present application scheme sets a smaller convolution kernel width according to different column correlations, thereby avoiding some features from being covered during weighting and calculation. At the same time, the convolution kernel length is set according to different time windows expressed in the data features, and the actual time series period of the data is considered. According to the periodic properties, at least three length sizes of convolution kernels are set to cover different features of short, medium and long time windows, so as to capture feature information of different time windows, and different widths and lengths of convolution kernels can be freely set according to different data performances to meet the feature collection requirements of various types of data. In addition, by sorting the features by relevance and using convolution kernels with smaller widths to process features with high relevance, more comprehensive feature information can be obtained.

[0081] According to an embodiment of the present application, an embodiment of a device failure prediction apparatus is also provided. Figure 7 Schematic diagram of a device failure prediction apparatus according to an embodiment of the present application. Figure 7 As shown, the device comprises:

[0082] The time series data acquisition module 70 is used to acquire the time series data of the running status of the device, wherein the time series data of the running status includes the index values ​​of multiple dimensions corresponding to the device at multiple data collection time points;

[0083] A convolution kernel design module 72, configured to determine the design parameters of the convolution kernel according to the first similarity and the second similarity corresponding to the operation status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index values ​​of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the operation status time series data is abnormal;

[0084] A feature extraction module 74 is used to construct a target convolution kernel according to the design parameters, and use the target convolution kernel to extract features from the operation status time series data to obtain target features;

[0085] The fault prediction module 76 is used to determine the fault risk probability of the equipment according to the target characteristics, wherein the fault risk probability is used to characterize the probability of the equipment failing in the future.

[0086] Optionally, the running status time series data contains multiple indicator data columns, each indicator data column is an indicator value arranged in time series order under one dimension, and the design parameters of the convolution kernel include: convolution kernel width and convolution kernel length; according to the first similarity and the second similarity corresponding to the running status time series data, determining the design parameters of the convolution kernel includes: determining the first similarity corresponding to each indicator data column in the running status time series data, and determining the convolution kernel width based on the first similarity; determining the second similarity between each indicator data column in the running status time series data and the abnormal point sequence, and determining the convolution kernel length based on the second similarity.

[0087] Optionally, based on the first similarity, determining the convolution kernel width includes: calculating the Pearson correlation coefficient between each pair of indicator data columns, and determining the Pearson correlation coefficient as the first similarity; determining the number of neighbors corresponding to each indicator data column, wherein the number of neighbors is the number of other indicator data columns whose first similarity with the indicator data column is greater than or equal to a preset similarity threshold; determining the minimum number of neighbors among the numbers of neighbors corresponding to each indicator data column, and when the value of the minimum number of neighbors is greater than 1, determining the value of the minimum number of neighbors as the convolution kernel width.

[0088] Optionally, the convolution kernel design module is also used to: select an indicator data column with the largest number of neighbors as the first column in the running status timing data; select a target number of indicator data columns from other indicator data columns that have not been selected in the running status timing data in order of the first similarity from large to small, and arrange them after the first column, where the target number is the value of the convolution kernel width minus 1; determine the last indicator data column that has been selected and sorted as the last column, and continue to select a target number of indicator data columns from other indicator data columns that have not been selected in order of the first similarity from large to small, until all the indicator data columns in the running status timing data have been selected and arranged.

[0089] Optionally, determining the convolution kernel length based on the second similarity includes: obtaining an outlier point sequence, wherein the outlier point sequence includes label values ​​corresponding to each data collection time point arranged in chronological order, wherein when the label value is a first numerical value, it represents that there is an anomaly in the indicator value corresponding to the data collection time point, and when the label value is a second numerical value, it represents that there is no anomaly in the indicator value corresponding to the data collection time point; calculating the Kendall correlation coefficient between the outlier point sequence and each indicator data column, and determining the Kendall correlation coefficient as the second similarity; determining the indicator data column with the largest absolute value of the second similarity in the operating status time series data as the benchmark data column; determining the period length corresponding to the benchmark data column, and determining the convolution kernel length based on the period length.

[0090] Optionally, the convolution kernel length includes: a baseline convolution kernel length, a first convolution kernel length, and a second convolution kernel length; determining the convolution kernel length based on the cycle length includes: performing a Fourier transform on the baseline data column to obtain frequency domain data corresponding to the baseline data column, wherein the Fourier transform is used to convert the baseline data column from the time domain to the frequency domain; determining the main frequency with the largest power in the frequency domain data, and determining the cycle length corresponding to the baseline data column based on the main frequency; determining the cycle length as the baseline convolution kernel length, and expanding the first convolution kernel length based on a first proportional coefficient and a second proportional coefficient to obtain a first convolution kernel length and a second convolution kernel length, wherein the first proportional coefficient is less than 1 and the second proportional coefficient is greater than 1.

[0091] Optionally, a target convolution kernel is used to extract features from the running status timing data to obtain target features, including: constructing a first convolution kernel based on the convolution kernel width and the first convolution kernel length, constructing a second convolution kernel based on the convolution kernel width and the second convolution kernel length, and constructing a third convolution kernel based on the convolution kernel width and the reference convolution kernel length; using the first convolution kernel, the second convolution kernel, and the third convolution kernel to respectively extract features from the running status timing data to obtain a first feature, a second feature, and a third feature; determining the target feature based on the first feature, the second feature, and the third feature.

[0092] It should be noted that the various modules in the above-mentioned equipment fault prediction device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0093] It should be noted that the equipment failure prediction device provided in this embodiment can be used to perform Figure 2 The device failure prediction method shown, therefore, the relevant explanations and descriptions of the above-mentioned device failure prediction method are also applicable to the embodiments of the present application and will not be repeated here.

[0094] An embodiment of the present application also provides a non-volatile storage medium, the non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the following device fault prediction method by running the computer program: obtaining the device's operating status time series data, wherein the operating status time series data includes index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; determining the design parameters of the convolution kernel based on the first similarity and the second similarity corresponding to the operating status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index values ​​of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the operating status time series data is abnormal; constructing a target convolution kernel according to the design parameters, and using the target convolution kernel to extract features from the operating status time series data to obtain target features; determining the failure risk probability of the device based on the target features, wherein the failure risk probability is used to characterize the probability of the device failing in the future.

[0095] The embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the device fault prediction method described in each embodiment of the present application: obtaining the operating status time series data of the device, wherein the operating status time series data contains index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; determining the design parameters of the convolution kernel based on the first similarity and the second similarity corresponding to the operating status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index values ​​of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the operating status time series data is abnormal; constructing a target convolution kernel according to the design parameters, and using the target convolution kernel to extract features from the operating status time series data to obtain target features; determining the failure risk probability of the device based on the target features, wherein the failure risk probability is used to characterize the probability of the device failing in the future.

[0096] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0097] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0099] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0102] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting equipment failure, characterized in that: include: Acquire the time series data of the running status of the device, wherein the time series data of the running status includes the index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; Determining the design parameters of the convolution kernel according to the first similarity and the second similarity corresponding to the running status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index value of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the running status time series data is abnormal; Constructing a target convolution kernel according to the design parameters, and using the target convolution kernel to extract features from the operating state time series data to obtain target features; The failure risk probability of the device is determined based on the target feature, wherein the failure risk probability is used to characterize the probability of the device failing in the future.

2. The equipment failure prediction method according to claim 1, characterized in that: The running state time series data includes a plurality of indicator data columns, each of which is an indicator value arranged in time series order under one dimension, and the design parameters of the convolution kernel include: convolution kernel width and convolution kernel length; according to the first similarity and the second similarity corresponding to the running state time series data, determining the design parameters of the convolution kernel includes: Determine the first similarity between each pair of the indicator data columns in the running status time series data, and determine the convolution kernel width according to the first similarity; Determine the second similarity between each of the indicator data columns in the operating status time series data and the abnormal point sequence, and determine the convolution kernel length based on the second similarity.

3. The equipment failure prediction method according to claim 2, characterized in that: Determining the convolution kernel width according to the first similarity includes: Calculating the Pearson correlation coefficient between each pair of the indicator data columns, and determining the Pearson correlation coefficient as the first similarity; Determine the number of nearest neighbors corresponding to each of the indicator data columns, wherein the number of nearest neighbors is the number of other indicator data columns whose first similarity with the indicator data column is greater than or equal to a preset similarity threshold; Determine the minimum number of neighbors among the neighbor numbers corresponding to each of the indicator data columns, and when the value of the minimum number of neighbors is greater than 1, determine the value of the minimum number of neighbors as the convolution kernel width.

4. The equipment failure prediction method according to claim 3, characterized in that: The method further comprises: Selecting the indicator data column with the largest number of neighbors as the first column in the running status time series data; In descending order of the first similarity with the first column, from other indicator data columns that are not selected from the running status time series data, a target number of the indicator data columns are selected and arranged after the first column, wherein the target number is the value of the convolution kernel width minus 1; The last indicator data column that has been selected and sorted is determined as the last column, and the target number of indicator data columns are selected from other indicator data columns that have not been selected and arranged in order of the first similarity with the last column from large to small, until all the indicator data columns in the operating status timing data have been selected and arranged.

5. The equipment failure prediction method according to claim 2, characterized in that: Determining the convolution kernel length according to the second similarity includes: Acquire the abnormal point sequence, wherein the abnormal point sequence includes label values ​​corresponding to each of the data collection time points arranged in chronological order, wherein when the label value is a first value, it indicates that the indicator value corresponding to the data collection time point is abnormal, and when the label value is a second value, it indicates that the indicator value corresponding to the data collection time point is not abnormal; Calculating the Kendall correlation coefficient between the abnormal point sequence and each of the indicator data columns, and determining the Kendall correlation coefficient as the second similarity; Determine the indicator data column with the largest absolute value of the second similarity in the operating status time series data as a reference data column; Determine the period length corresponding to the reference data column, and determine the convolution kernel length based on the period length.

6. The equipment failure prediction method according to claim 5, characterized in that: The convolution kernel length includes: a reference convolution kernel length, a first convolution kernel length, and a second convolution kernel length; and determining the convolution kernel length according to the period length includes: Performing Fourier transform on the reference data sequence to obtain frequency domain data corresponding to the reference data sequence, wherein the Fourier transform is used to convert the reference data sequence from the time domain to the frequency domain; Determine a main frequency with the largest power in the frequency domain data, and determine the period length corresponding to the reference data sequence based on the main frequency; The period length is determined as the benchmark convolution kernel length, and the first convolution kernel length is expanded according to a first proportional coefficient and a second proportional coefficient to obtain the first convolution kernel length and the second convolution kernel length, wherein the first proportional coefficient is less than 1 and the second proportional coefficient is greater than 1.

7. The equipment failure prediction method according to claim 6, characterized in that: The target convolution kernel is used to extract features from the operating status time series data, and the target features obtained include: Constructing a first convolution kernel according to the convolution kernel width and the first convolution kernel length, constructing a second convolution kernel according to the convolution kernel width and the second convolution kernel length, and constructing a third convolution kernel according to the convolution kernel width and the reference convolution kernel length; Using the first convolution kernel, the second convolution kernel, and the third convolution kernel, respectively, to extract features from the operating status time series data to obtain a first feature, a second feature, and a third feature; The target feature is determined based on the first feature, the second feature, and the third feature.

8. An equipment failure prediction device, characterized in that: include: A time series data acquisition module, used to acquire the time series data of the running status of the device, wherein the time series data of the running status includes the index values ​​of multiple dimensions corresponding to the device at multiple data collection time points; A convolution kernel design module, used to determine the design parameters of the convolution kernel according to the first similarity and the second similarity corresponding to the running status time series data, wherein the first similarity is used to characterize the correlation between the index values ​​of each dimension, and the second similarity is used to characterize the correlation between the index value of each dimension and the abnormal point sequence, and the abnormal point sequence is used to indicate the data collection time point at which the index value in the running status time series data is abnormal; A feature extraction module, used to construct a target convolution kernel according to the design parameters, and use the target convolution kernel to extract features from the operating state time series data to obtain target features; A fault prediction module is used to determine the fault risk probability of the device based on the target feature, wherein the fault risk probability is used to characterize the probability of the device failing in the future.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the device failure prediction method according to any one of claims 1 to 7 by running the computer program.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the equipment failure prediction method according to any one of claims 1 to 7 are implemented.

Images