Data processing method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a data processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring a first time sequence and a second time sequence; the first time sequence comprises a plurality of first data sets; the second time sequence comprises a plurality of second data sets; based on a dynamic regular path between the first time sequence and the second time sequence, determining a first target data set from each first data set and a second target data set corresponding to the first target data set from each second data set; performing dimension reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time sequence, and performing dimension reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time sequence; and performing kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time sequence and the second time sequence. By adopting the method, the accuracy of the correlation analysis result can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of big data processing, and in particular to a data processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art

[0002] As the amount of data grows, more complex multivariate relationships are formed between data sets. How to efficiently mine the implicit information behind the data is of great significance. Correlation analysis is one of the important technologies to simplify data and explore hidden relationships.

[0003] Generally, the correlation between data sets can be mined based on local correlation analysis, but when analyzing the correlation between high-dimensional data sets, the correlation analysis results are inaccurate. Summary of the invention

[0004] Based on this, it is necessary to provide a data processing method, apparatus, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of association analysis results in response to the above technical problems.

[0005] In the first aspect, the present application provides a data processing method. The method includes: obtaining a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; based on the dynamic regularization path between the first time series sequence and the second time series sequence, determining a first target data set from each first data set, and a second target data set corresponding to the first target data set in each second data set; performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result of the first time series sequence and the second time series sequence.

[0006] In one embodiment, a kernel canonical correlation analysis is performed on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result of the first time series sequence and the second time series sequence, including: obtaining a first weight feature that matches the first comprehensive feature and a second weight feature that matches the second comprehensive feature; based on the first weight feature, the second weight feature and the Gaussian kernel function, a kernel canonical correlation analysis is performed on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result of the first time series sequence and the second time series sequence.

[0007] In one embodiment, the first comprehensive feature includes multiple first sub-features; the method also includes: obtaining multiple first candidate weight features; for each first candidate weight feature, weighted summing the first candidate weight feature to each first sub-feature to determine the first indicator feature corresponding to the first candidate weight feature; the first indicator feature characterizes the degree of aggregation of valid features in the first time series; based on the redundant analysis results of the first comprehensive features matching each first indicator feature, the first candidate weight feature corresponding to the largest redundant analysis result is determined as the first weight feature matching the first comprehensive feature.

[0008] In one embodiment, there are multiple first weight features, each of which includes multiple first sub-weights. The method further includes: based on the first sub-weights corresponding to each first sub-feature in each first weight feature, mapping each first sub-feature and the first indicator feature corresponding to each first weight feature, determining and visually displaying the first mapping result using a Sankey diagram; wherein the first mapping result includes multiple first sub-mapping results, and the first sub-mapping result represents the importance of the first sub-feature in the first comprehensive feature; wherein the first node in the visually displayed Sankey diagram represents each first sub-feature in the first comprehensive feature, and the second node in the visually displayed Sankey diagram represents the first indicator feature corresponding to each first weight feature; the edge between the first node and the second node represents each first sub-weight corresponding to the first sub-feature; and the indicator value on the first indicator feature is the sum of the first sub-weights corresponding to each mapped first sub-feature.

[0009] In one embodiment, the method further comprises: determining a redundancy analysis result of the first comprehensive feature matching the first indicator feature based on a ratio between the variance of the first indicator feature and the total variance of the first comprehensive feature.

[0010] In one embodiment, there are multiple first target data sets, each of which includes multiple first target features; a dimensionality reduction analysis is performed on the first target data sets to determine the first comprehensive feature corresponding to the first time series, including: for each first target feature in each first target data set, a dimensionality reduction process is performed on each first target feature in the first target data set to obtain the dimensionality reduction feature corresponding to the first target data set; and the dimensionality reduction features corresponding to each first target data set are combined to obtain the first comprehensive feature corresponding to the first time series.

[0011] In a second aspect, the present application provides a data processing device, which includes: an acquisition module for acquiring a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; a determination module for determining a first target data set from each first data set and a second target data set corresponding to the first target data set in each second data set based on a dynamically regularized path between the first time series sequence and the second time series sequence; an analysis module for performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; a processing module for performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result between the first time series and the second time series.

[0012] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program: obtaining a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; based on the dynamic regularization path between the first time series sequence and the second time series sequence, determining a first target data set from each first data set, and a second target data set corresponding to the first target data set in each second data set; performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result between the first time series and the second time series.

[0013] In a fourth aspect, the present application also provides a computer-readable storage medium. A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented: obtaining a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; based on the dynamic regularization path between the first time series sequence and the second time series sequence, determining a first target data set from each first data set, and a second target data set corresponding to the first target data set in each second data set; performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result between the first time series and the second time series.

[0014] In a fifth aspect, the present application also provides a computer program product. The computer program product includes a computer program, which implements the following steps when executed by a processor: obtaining a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; based on the dynamic regularization path between the first time series sequence and the second time series sequence, determining a first target data set from each first data set, and a second target data set corresponding to the first target data set in each second data set; performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result between the first time series and the second time series.

[0015] The above-mentioned data processing method, device, computer equipment, storage medium and computer program product obtain a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets, and based on the dynamic regularization path between the first time series sequence and the second time series sequence, determine the first target data set from each first data set, and the second target data set corresponding to the first target data set in each second data set, and perform dimensionality reduction analysis on the first target data set to determine the first comprehensive feature corresponding to the first time series sequence, and perform dimensionality reduction analysis on the second target data set to determine the second comprehensive feature corresponding to the second time series sequence, and then perform kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain the association analysis result of the first time series sequence and the second time series sequence. Therefore, by performing kernel canonical correlation analysis, the nonlinear association relationship between data can be mined in high-dimensional space, avoiding the inaccurate association analysis result caused by ignoring the nonlinear trend between different data sets, and improving the accuracy of the association analysis result. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 An application environment diagram of a data processing method in an embodiment;

[0017] Figure 2 is a flow chart of a data processing method in one embodiment;

[0018] Figure 3 A schematic diagram of a dynamically regularized path in one embodiment;

[0019] Figure 4 A schematic diagram of correlation analysis in one embodiment;

[0020] Figure 5It is a flowchart of performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time series sequence and the second time series sequence in one embodiment;

[0021] Figure 6 is a flow chart of a data processing method in another embodiment;

[0022] Figure 7 is a flow chart of a data processing method in another embodiment;

[0023] Figure 8 is a schematic diagram of a visualization result in one embodiment;

[0024] Fig. 9 is a flow chart of a data processing method in another embodiment;

[0025] Fig.10 is a structural block diagram of a data processing device in one embodiment;

[0026] Fig.11 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0028] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations; and the acquisition, storage, processing, transmission, etc. of data must comply with relevant laws and regulations.

[0029] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0030] Correlation analysis is also called association analysis, which refers to the study of the closeness of the correlation between two or two groups of variables. Common analysis techniques are mainly divided into local correlation analysis and global correlation analysis. At present, some researchers have conducted in-depth research on local correlation. For example, the Pearson correlation coefficient is a commonly used indicator for studying local correlation. It is usually used to reflect the linear correlation between two parameters X and Y. Its calculation formula is as follows:

[0031]

[0032] Pearson correlation coefficient The size of is the covariance of parameters X and Y divided by their standard deviation and The product of and represent the means of parameters X and Y respectively, The value range of is [-1,1]. , it means there is a positive correlation between parameters X and Y; if , it means that there is a negative correlation between parameters X and Y; if , then it means that there is no correlation between parameters X and Y.

[0033] However, the Pearson correlation coefficient is generally used to calculate the linear correlation between two sets of data, while in the field of big data, data sets often show nonlinear trends. Therefore, if the Pearson correlation coefficient is continued to be used to evaluate the correlation between two parameters, the final result may be inaccurate. Therefore, when outliers appear in the data or there is a nonlinear relationship, the Spearman method is more suitable. It is to perform linear correlation analysis by using the rank size of two sets of variables, which is equivalent to calculating the Pearson correlation coefficient of two sets of sorted variables.

[0034] Assume that both sets of parameters X and Y are n-dimensional sets, record , First, sort the set in ascending or descending order, and take the sorted set elements as , , then the Spearman calculation formula is as follows:

[0035]

[0036] in, and It represents the mean of sets X and Y. The value of the Spearman correlation coefficient is also [-1,1]. The larger its absolute value, the stronger the correlation between sets X and Y.

[0037] Although Spearman can measure the local correlation between two data sets, it will produce redundant pairwise correlations, which brings challenges to rapid data analysis. Canonical Correlation Analysis (CCA) focuses on studying the global correlation between parameter sets. CCA starts from the overall correlation structure between parameter sets, extracts linear combinations of several groups of variables from two sets of variables, namely comprehensive indicators, and measures the correlation between the original parameter sets by studying the correlation between comprehensive indicators. First, assume that there are two sets of time series data sets and The lengths of the two time series are p and q, which are also expressed as the number of parameters. The purpose of CCA is to mine typical variables (i.e., comprehensive indicators) from the original data set, and then find each set of original variables and CCA will produce d pairs of canonical variables, where , the typical variable is recorded as and The typical variable represents the linear combination of the original variables, and the specific formula is as follows:

[0038]

[0039]

[0040] in, and As the parameter coefficient of the linear combination, it is also called the typical weight. Its size indicates the importance of the original variable to the typical variable and is an important indicator to measure the contribution of the parameter.

[0041] The main goal of CCA is to find the typical variables and The maximum correlation between :

[0042]

[0043] Where C represents covariance. Although CCA can mine the relationship between two groups of variables, it assumes that there is a linear relationship between the two groups of variables, and its main goal is to calculate the linear correlation between variables, ignoring the nonlinear trend between data sets. Therefore, the analysis results of CCA may have some errors.

[0044] In order to solve the problem of nonlinear data sets, the Kernel Canonical Correlation Analysis (KCCA) method can be used. KCCA introduces the kernel function on the basis of linear CCA, maps the data set to a high-dimensional feature space through the kernel function, and then mines the nonlinear correlation between data in the high-dimensional space. In other words, by finding the maximum correlation of the kernel canonical variables, the correlation analysis between data sets can be achieved.

[0045] From the above, we can see that although local correlation analysis can mine the correlation between data, when analyzing high-dimensional data sets, redundant binary correlations will be generated, thus increasing the burden of insight results. When studying the global correlation between many-to-many data sets, CCA has better analysis results. Although CCA is a powerful method to associate the relationship between two multidimensional variables, the main goal of CCA is to find the linear relationship between two sets of data sets. When there is a nonlinear correlation between the two variables, the analysis results of linear CCA may produce partial errors. KCCA is a common method to solve the problem of CCA in processing nonlinear data. It maps the original data set to a high-dimensional feature space through a kernel function, and then performs linear CCA in the new feature space.

[0046] At present, most studies focus on KCCA to solve the nonlinear correlation between single data sets, and cannot measure the correlation between high-dimensional multivariate complex data sets. In addition, due to the principle of KCCA, multiple pairs of core typical correlations will be generated, and each pair of correlations has many-to-many correlations, thus forming complex organizational characteristics, increasing the burden of user insight results, and bringing huge challenges to its research and analysis. In recent years, visual analysis has combined data visualization methods to integrate expert knowledge into the data analysis process, improving the flexibility, interpretability and accuracy of the analysis process, and has become an indispensable key technology in the field of data analysis.

[0047] In combination with the above analysis, this application provides a data processing method, which can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 through a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers.

[0048] Specifically, the server 104 obtains a first time series sequence and a second time series sequence from the terminal 102; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; and then the server 104 determines a first target data set from each first data set and a second target data set corresponding to the first target data set in each second data set based on the dynamic regularization path between the first time series sequence and the second time series sequence, and performs a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performs a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence, and then, by performing a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature, the association analysis result of the first time series sequence and the second time series sequence can be obtained.

[0049] The terminal 102 may be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers.

[0050] In one embodiment, Figure 2 As shown, a data processing method is provided, which is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0051] S202, acquiring a first time series sequence and a second time series sequence; the first time series sequence includes a plurality of first data sets; the second time series sequence includes a plurality of second data sets.

[0052] In this embodiment, each first data set includes multiple first parameter values ​​corresponding to the first target parameter, each second data set includes multiple second parameter values ​​corresponding to the second target parameter, and each first data set and each second data set have the characteristics of a high-dimensional time series. The first target parameter and the second target parameter refer to the operating parameters of different associated devices, and the multiple first parameter values ​​refer to the parameter values ​​of the first target parameter collected at different sampling times, and the multiple second parameter values ​​refer to the parameter values ​​of the second target parameter collected at different sampling times. The form of the first target parameter and the second target parameter is not limited. For example, the first target parameter refers to the voltage of the battery, and the second target parameter refers to the temperature of the battery.

[0053] For example, the first time series , that is, the number of first data sets is m; for any first data set in the first time series L , , s can be the same as or different from m. The second timing sequence , that is, the number of second data sets is n; for any second data set in the second time series sequence T , , r may be the same as or different from n.

[0054] In some embodiments, the first time series sequence is data obtained by preprocessing the first original data set, and the second time series sequence is data obtained by preprocessing the second original data set.

[0055] The way of preprocessing the first original data set and the way of preprocessing the second original data set may be the same or different. For example, the original data set may be preprocessed using a min-max or Z-Score normalization method.

[0056] S204 , based on the dynamic regularization path between the first time series sequence and the second time series sequence, determine a first target data set from each of the first data sets and a second target data set corresponding to the first target data set from each of the second data sets.

[0057] There is no limitation on the method for obtaining the dynamically regularized path between the first timing sequence and the second timing sequence, and a description is given below in combination with possible implementation methods.

[0058] In one embodiment, a dynamic time warping (DTW) method may be used to obtain a dynamic warping path between the first timing sequence and the second timing sequence.

[0059] Specifically, based on the element distances between the elements in each first data set and the elements in each second data set, a distance matrix is ​​obtained, and each distance in the distance matrix is ​​mapped to the target grid; when the reference distance is determined from the target grid, the position of the reference distance in the target grid is used as the initial position, and the minimum distance is determined from the distances adjacent to the reference distance, and the minimum distance is used as the new reference distance, and the step of taking the position of the reference distance in the target grid as the initial position is returned until there is no adjacent distance for the new reference distance; based on the positions of multiple minimum distances in the target grid, a dynamic regularization path between the first time series and the second time series is determined. The dynamic regularization path refers to the shortest similarity path between the first data set and the second data set.

[0060] In some cases, the reference distance may be the distance corresponding to the lower left grid in the target grid, and the iterative process is stopped when the new reference distance is the distance corresponding to the upper right grid in the target grid. In other cases, the reference distance may be the distance corresponding to the upper right grid in the target grid, and the iterative process is stopped when the new reference distance is the distance corresponding to the lower left grid in the target grid.

[0061] For example, the reference distance can be the distance corresponding to the lower left corner of the target grid. Figure 3As shown in Figure 1, a schematic diagram of a dynamic regularization path is provided. In which, each grid corresponds to an element distance, then the dynamic regularization path , W k Represents the regularized path corresponding to the Kth element of the alignment between the first time series sequence and the second time series sequence.

[0062] Combination Figure 3 The target network shown in the figure has only three directions for each grid path according to the dynamic regularization optimization criterion. For example, if the path has passed through the grid , then with the grid Adjacent grids, i.e. grids The next grid can only be , or Under the premise of meeting the above conditions, the shortest similarity path between the first time series L and the second time series T can be achieved. The distance corresponding to the current grid In the case of the sum of , the calculation formula is as follows:

[0063]

[0064] Thus, based on the minimum path function , we can get Figure 3 A dynamically regularized path W between the first timing sequence and the second timing sequence is shown.

[0065] Furthermore, based on the dynamically regularized path between the first time series sequence and the second time series sequence, the first data set and the second data set corresponding to each path point in the dynamically regularized path are determined, thereby determining the first data set corresponding to each path point as the first target data set, that is, the number of first target data sets is consistent with the number of path points in the dynamically regularized path; similarly, the second data set corresponding to each path point is determined as the second target data set, that is, the number of second target data sets is consistent with the number of path points in the dynamically regularized path.

[0066] For example, combining Figure 3 , the path points in the dynamic regularization path are to ,in, The corresponding first data set is , the corresponding second data set is ; The corresponding first data set is , the corresponding second data set is ; The corresponding first data set is , the corresponding second data set is , and so on, a first target data set and a second target data set can be obtained, the first target data set includes a first data set corresponding to each path point, and the second target data set includes a second data set corresponding to each path point.

[0067] S206, performing dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence.

[0068] The first target data set includes multiple first target features. For example, Figure 3 , dynamically regularize the waypoints in the path The corresponding first data set is ,but is the first target data set, the first target data set Corresponding The first target dataset Multiple first target features in .

[0069] The number of the first target data sets may be one or more. There is no limitation on the implementation method of performing dimensionality reduction analysis on the first target data set to determine the first comprehensive feature corresponding to the first time series sequence, and several of the methods are described below by way of example.

[0070] In one embodiment, there are multiple first target data sets. By inputting each first target feature in the first target data set into the dimensionality reduction model, the dimensionality reduction feature corresponding to the first target data set can be obtained; the dimensionality reduction features corresponding to each first target data set are combined to obtain the first comprehensive feature corresponding to the first time series.

[0071] In one embodiment, there are multiple first target data sets, and for each first target feature in each first target data set, dimensionality reduction processing is performed on each first target feature in the first target data set to obtain the reduced dimensionality feature corresponding to the first target data set; the reduced dimensionality features corresponding to each first target data set are combined to obtain the first comprehensive feature corresponding to the first time series.

[0072] The principal component analysis (PCA) method may be used to perform dimensionality reduction processing on each first target feature in the first target data set, or other dimensionality reduction methods may be used to perform dimensionality reduction processing.

[0073] Among them, the dimensionality reduction analysis of the second target data set is performed to determine the implementation method of the second comprehensive feature corresponding to the second time series sequence. Reference can be made to the dimensionality reduction analysis of the first target data set to determine the implementation method of the first comprehensive feature corresponding to the first time series sequence, or other methods can also be used.

[0074] S208, performing kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time series sequence and the second time series sequence.

[0075] Among them, Kernel Canonical Correlation Analysis (KCCA) introduces the kernel function on the basis of CCA. The corresponding kernel function can be selected according to the characteristics of the data set. Thus, the first comprehensive feature and the second comprehensive feature are mapped to the high-dimensional feature space through the kernel function, and then the nonlinear correlation relationship between the first comprehensive feature and the second comprehensive feature is mined in the high-dimensional space to obtain the correlation analysis results of the first time series and the second time series.

[0076] The kernel function may include but is not limited to: linear kernel function, Gaussian kernel function and multilayer perceptron (MLP) kernel function. Figure 4 As shown, a correlation analysis schematic diagram is provided. Among them, the number of first target data sets is q, and the number of second target data sets is p. By using the PCA method to reduce the dimension of each first target feature in each first target data set, the reduced dimension features corresponding to each first target data set can be obtained. By combining the reduced dimension features corresponding to each first target data set, the first comprehensive feature can be obtained. Similarly, the second comprehensive feature can be obtained. Furthermore, the KCCA method is used to perform correlation analysis on the first comprehensive feature and the second comprehensive feature, and the association analysis results of the first time series sequence and the second time series sequence can be obtained.

[0077] based on Figure 2 The content shown is obtained by obtaining a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets, and based on the dynamic regularization path between the first time series sequence and the second time series sequence, the first target data set is determined from each first data set, and the second target data set corresponding to the first target data set in each second data set, and the first target data set is subjected to dimensionality reduction analysis to determine the first comprehensive feature corresponding to the first time series sequence, and the second target data set is subjected to dimensionality reduction analysis to determine the second comprehensive feature corresponding to the second time series sequence, and then the first comprehensive feature and the second comprehensive feature are subjected to kernel typical correlation analysis to obtain the association analysis result of the first time series sequence and the second time series sequence. Therefore, by performing kernel typical correlation analysis, the nonlinear association relationship between data can be mined in high-dimensional space, avoiding the inaccurate association analysis result caused by ignoring the nonlinear trend between different data sets, and improving the accuracy of the association analysis result.

[0078] In one embodiment, Figure 5 As shown, a flowchart of performing a kernel canonical correlation analysis on a first comprehensive feature and a second comprehensive feature to obtain a correlation analysis result of a first time series sequence and a second time series sequence is provided, including the following steps:

[0079] S502, obtaining a first weighted feature matching the first comprehensive feature and a second weighted feature matching the second comprehensive feature.

[0080] There is no limitation on the implementation method of determining the first weight feature that matches the first comprehensive feature and the second weight feature that matches the second comprehensive feature. The following is an example of possible implementation methods.

[0081] In one embodiment, the first preset weight feature may be determined as the first weight feature that matches the first comprehensive feature; and the second preset weight feature may be determined as the second weight feature that matches the second comprehensive feature.

[0082] S504: Based on the first weight feature, the second weight feature and the Gaussian kernel function, a kernel canonical correlation analysis is performed on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time series sequence and the second time series sequence.

[0083] Specifically, the correlation analysis results of the first time series and the second time series are satisfy:

[0084]

[0085] in, represents the Gaussian kernel function, , x and y represent the first comprehensive feature and the second comprehensive feature respectively, represents the bandwidth parameter, which is used to control the local range of the Gaussian kernel function. The larger the value of , the larger the local influence range of the Gaussian kernel function; conversely, The smaller it is, the smaller the local influence range of the Gaussian kernel function is. represents the first weight feature, Represents the second weight feature.

[0086] Taking the correlation analysis method using KCCA and CCA, and the kernel function in KCCA being the Gaussian kernel function and the linear kernel function as an example, by performing correlation analysis on the first eight groups of data between different test data sets, the association analysis results shown in Table 1 are obtained.

[0087]

[0088] As can be seen from Table 1, when the CCA method is used for analysis, the correlations between the first eight groups of data are relatively high, all above 0.6. CCA is based on the result obtained when both groups of data are in a linear relationship. Therefore, the correlations between the first eight groups of data are relatively strong, with a large error, and cannot truly reflect the correlation between the two groups of data sets.

[0089] KCCA has a better discriminant effect when solving nonlinear data sets, but the results of different kernel functions are also different. For example, in Table 1, for KCCA using the Gaussian kernel function, the correlation between the first set of data is high, and then decreases successively and changes regularly. For KCCA using the linear kernel function, the correlation changes are irregular, and the data changes are relatively scattered; the linear kernel is suitable for objects represented by a large number of fixed-length features, and it is assumed that they are linearly separable in space. Therefore, after comparing the analysis results, when this application uses the Gaussian kernel function for correlation analysis, the accuracy of the obtained association analysis results can be improved.

[0090] In one embodiment, the first comprehensive feature includes a plurality of first sub-features, specifically, Figure 6 As shown, a data processing method is provided, which is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0091] S602: Obtain multiple first candidate weight features.

[0092] Among them, multiple first candidate weight features can be obtained based on the covariance corresponding to the first comprehensive feature.

[0093] Specifically, based on the feature mean of each first sub-feature in the first comprehensive feature, each first sub-feature in the first comprehensive feature data is standardized to obtain the first standard feature; the covariance feature corresponding to the first standard feature is subjected to eigenvalue decomposition, and each feature vector obtained after the decomposition is determined as each first candidate weight feature. That is, the number of feature vectors is the same as the number of first candidate weight features.

[0094] The first comprehensive feature may be a one-dimensional feature or a multi-dimensional feature. When the first comprehensive feature is a multi-dimensional feature, each first sub-feature in the first comprehensive feature refers to multiple first sub-features of the same dimension, and the corresponding first standard feature can be obtained for each of the multiple first sub-features of each dimension.

[0095] S604: For each first candidate weight feature, perform weighted summation of the first candidate weight feature on each first sub-feature to determine a first indicator feature corresponding to the first candidate weight feature.

[0096] Among them, the first candidate weight feature includes first candidate coefficients corresponding to each first sub-feature. By weighted summing the first candidate coefficients corresponding to each first sub-feature to the corresponding first sub-feature, the first indicator feature corresponding to the first candidate weight feature can be determined, and the first indicator feature represents the degree of aggregation of effective features in the first comprehensive feature.

[0097] For example, the first indicator feature satisfy: . Where X represents the first comprehensive feature, represents each first sub-feature in the first comprehensive feature X, represents the first candidate weight feature, Represent the first candidate weight features respectively The first candidate coefficients corresponding to each first sub-feature in .

[0098] Among them, when the first comprehensive feature is a multidimensional feature, Can be used to represent the first sub-features corresponding to each dimension, then the first indicator feature It is also a multidimensional feature.

[0099] S606: Based on the redundant analysis results of the first comprehensive features matching the first indicator features, determine the first candidate weight feature corresponding to the largest redundant analysis result as the first weight feature matching the first comprehensive feature.

[0100] There is no limitation on the method for determining the redundant analysis result of the first comprehensive feature matching the first indicator feature, and several of the methods are described below by way of example.

[0101] In one embodiment, the first indicator feature is input into a redundancy analysis model to obtain a redundancy analysis result of a first comprehensive feature matching the first indicator feature, wherein the redundancy analysis model is trained based on the redundancy analysis results of each indicator feature and the corresponding comprehensive feature.

[0102] In one embodiment, based on the ratio between the variance of the first indicator feature and the total variance of the first comprehensive feature, the redundant analysis result of the first comprehensive feature matching the first indicator feature is determined. In some cases, the ratio between the variance of the first indicator feature and the total variance of the first comprehensive feature can be determined as the redundant analysis result of the first comprehensive feature matching the first indicator feature.

[0103] Among them, the redundancy analysis result represents the proportion of the total variance in the first comprehensive feature that can be explained by the first indicator feature. Therefore, by quantitatively measuring the amount of original information contained in the first indicator feature, it is helpful to understand to what extent the first indicator feature can represent the first comprehensive feature, so as to more accurately grasp the intrinsic relationship between the first time series and the second time series.

[0104] The implementation method of determining the second weight feature matching the second comprehensive feature can refer to Figure 6 The content is suitable for the description and will not be repeated here.

[0105] Further, when a first weight feature matching the first comprehensive feature and a second weight feature matching the second comprehensive feature are determined based on the maximum redundancy analysis result, the first weight feature and the second weight feature are substituted into the association analysis results of the first time series and the second time series. The satisfied formula can obtain the correlation analysis result of the first time series and the second time series.

[0106] based on Figure 6 The method shown can determine the first weight feature that matches the first comprehensive feature from the perspective of redundant analysis. Therefore, when obtaining the correlation analysis results of the first time series sequence and the second time series sequence based on the obtained first weight feature, it can avoid the inaccurate analysis results caused by data redundancy and improve the accuracy of the analysis results.

[0107] It should be understood that after using KCCA for association analysis, multiple pairs of typical core association relationships will be generated, and each pair of association relationships has many-to-many correlations, thus forming complex organizational features and increasing the burden on users to gain insight into the results. Although the traditional node link-oriented graph can visualize the association characteristics between multiple parameters, when there are many parameters, it is easy to cause node occlusion. Although the ring relationship graph can overcome the node occlusion problem, its layout is relatively chaotic and it is not easy to gain insight into the association between nodes. The Sankey diagram shows the flow of data through the composition of edges, flows and fulcrums. It can not only describe the flow characteristics of the data but also visualize the association relationship, and has a clear and intuitive visualization effect. Therefore, the relevant analysis results can be visualized based on the Sankey diagram, so that based on the visualization results and integrating the user's domain knowledge into the visual analysis process, the flexibility and interpretability of the analysis process can be increased to achieve efficient analysis.

[0108] In one embodiment, the number of first weight features is multiple, each first weight feature includes multiple first sub-weights, and the number of first indicator features is also multiple. Similarly, the number of second weight features is multiple, each second weight feature includes multiple second sub-weights, and the number of second indicator features is multiple.

[0109] Specifically, Figure 7 As shown, a data processing method is provided, which is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0110] S702, based on the first sub-weights corresponding to the first sub-features in the first weight features, map the first sub-features to the first indicator features corresponding to the first weight features, determine and visualize the first mapping results using a Sankey diagram.

[0111] The first mapping result includes a plurality of first sub-mapping results, and the first sub-mapping result represents the importance of the first sub-feature in the first comprehensive feature.

[0112] Among them, the first node in the visualized Sankey diagram represents each first sub-feature in the first comprehensive feature, and the second node in the visualized Sankey diagram represents the first indicator feature corresponding to each first weight feature; the edge between the first node and the second node represents each first sub-weight corresponding to the first sub-feature.

[0113] Among them, the index value on the first index feature is the sum of the first sub-weights corresponding to the mapped first sub-features, maintaining energy conservation and following the basic principle of the Sankey diagram.

[0114] S704: Based on the second sub-weights corresponding to the second sub-features in the second weight features, each second sub-feature is mapped to a second indicator feature corresponding to each second weight feature, and a second mapping result is determined and visualized using a Sankey diagram.

[0115] The second mapping result includes a plurality of second sub-mapping results, and the second sub-mapping results represent the importance of the second sub-feature in the second comprehensive feature.

[0116] Among them, the third node in the visualized Sankey diagram represents each second sub-feature in the second comprehensive indicator, and the fourth node in the visualized Sankey diagram represents the second indicator feature corresponding to each second weight feature; the edge between the third node and the fourth node represents the second sub-weights corresponding to the second sub-feature; the indicator value on the second indicator feature is the sum of the second sub-weights corresponding to the mapped second sub-features.

[0117] Based on the contents shown in S702-S704, taking the test data set mentioned in Table 1 as an example, Figure 8 As shown, a schematic diagram of a visualization result is provided. Among them, UV1~UV3 respectively represent the correlation analysis results of the first time series and the second time series, the first node represents each first sub-feature in the first comprehensive feature X, the second node U1~U3 respectively represents each first indicator feature, the third node represents each second sub-feature in the second comprehensive feature X, and the fourth node V1~V3 respectively represents each second indicator feature.

[0118] exist Figure 8In the figure, UV1=0.99, the first weight feature used to obtain UV1 is the same as the first weight feature used to obtain the first indicator feature U1, and the second weight feature used to obtain UV1 is the same as the second weight feature used to obtain the second indicator feature V1. UV2=0.82, the first weight feature used to obtain UV2 is the same as the first weight feature used to obtain the first indicator feature U2, and the second weight feature used to obtain UV2 is the same as the second weight feature used to obtain the second indicator feature V2. UV3=0.61, the first weight feature used to obtain UV3 is the same as the first weight feature used to obtain the first indicator feature U3, and the second weight feature used to obtain UV3 is the same as the second weight feature used to obtain the second indicator feature V3.

[0119] based on Figure 8 It can be seen that when the number of the first indicator features and the second indicator features are 3 respectively, each first sub-feature in the first comprehensive feature needs to be mapped to the second nodes U1~U3. Similarly, each second sub-feature in the second comprehensive feature needs to be mapped to the third nodes V1~V3.

[0120] exist Figure 8 In the figure, the value 0.35 corresponding to "U1-X" is used to characterize the redundant analysis result of the first comprehensive feature X matching the first indicator feature U1; the value 0.37 corresponding to "V1-Y" is used to characterize the redundant analysis result of the second comprehensive indicator Y matching the second indicator feature V1.

[0121] based on Figure 8 As shown in the content, in some cases, a legend function can be set on the visualized Sankey diagram to display the correlation analysis results corresponding to each group of indicator characteristics; it can also be used to achieve interactive linkage by clicking on the legend in the Sankey diagram, focusing on analyzing the correlation characteristics under different core typical correlation relationships.

[0122] In combination with the above, in one embodiment, Fig. 9 As shown, a data processing method is provided, which is applied to Figure 1 Taking the server 104 in the example as an example, the following steps are included:

[0123] S902, acquiring a first time series sequence and a second time series sequence; the first time series sequence includes a plurality of first data sets; the second time series sequence includes a plurality of second data sets.

[0124] S904 , based on the dynamic regularization path between the first time series sequence and the second time series sequence, determine a first target data set from each of the first data sets and a second target data set corresponding to the first target data set from each of the second data sets.

[0125] There are multiple first target data sets and multiple second target data sets, each first target data set includes multiple first target features, and each second target data set includes multiple second target features.

[0126] 906. For each first target feature in each first target data set, use the PCA method to perform dimensionality reduction processing on each first target feature in the first target data set to obtain the reduced dimensionality feature corresponding to the first target data set; combine the reduced dimensionality features corresponding to each first target data set to obtain the first comprehensive feature corresponding to the first time series.

[0127] S908, based on the feature mean of each first sub-feature in the first comprehensive feature, standardize each first sub-feature in the first comprehensive feature data to obtain a first standard feature; perform eigenvalue decomposition on the covariance feature corresponding to the first standard feature, and determine each feature vector obtained after the decomposition as each first candidate weight feature.

[0128] S910: For each first candidate weight feature, perform weighted summation of the first candidate weight feature to each first sub-feature to determine a first indicator feature corresponding to the first candidate weight feature.

[0129] S912: Determine the ratio between the variance of the first indicator feature corresponding to the first candidate weight feature and the total variance of the first comprehensive feature as the redundancy analysis result of the first comprehensive feature matching the first indicator feature.

[0130] S914: Based on the redundant analysis results of the first comprehensive features matching the first indicator features, determine the first candidate weight feature corresponding to the largest redundant analysis result as the first weight feature matching the first comprehensive feature.

[0131] S916, determining a second weighted feature that matches the second comprehensive feature.

[0132] Among them, the implementation method of determining the second weight feature that matches the second comprehensive feature can refer to the content adaptation description of S908-S914, which will not be repeated here.

[0133] S918, based on the first weight feature, the second weight feature and the Gaussian kernel function, perform kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time series sequence and the second time series sequence.

[0134] Among them, the correlation analysis results of the first time series and the second time series satisfy: .in, represents the Gaussian kernel function, , x and y represent the first comprehensive feature and the second comprehensive feature respectively, represents the bandwidth parameter, represents the first weight feature, Represents the second weight feature.

[0135] Among them, the specific content of S902-S918 can be adapted to the description of the above content and will not be repeated here.

[0136] based on Fig. 9 As shown in the content, when determining the correlation analysis result of the first time series and the second time series, the redundant analysis result corresponding to the weight feature used is the largest, such as Figure 8 UV1=0.99 in the equation, therefore, data mining and data analysis can be performed based on the target data corresponding to the maximum redundant analysis result. The target data may include each first sub-feature and each second sub-feature corresponding to the maximum redundant analysis result, and the first weight feature and the second weight feature corresponding to the maximum redundant analysis result.

[0137] based on Fig. 9 As shown in the content, firstly, in order to solve the problem of inconsistent step lengths and calculation errors caused by the time series data, the dynamic time warping algorithm is used to determine the corresponding dynamic warping path. Then, based on the high-dimensional multivariate nature of parameter features, the data of the same element of different dimensions in the parameter set after dynamic warping are combined, and then the principal component analysis method is used to extract features of each element, and the comprehensive representative features (i.e., the first comprehensive feature and the second comprehensive feature) between the data sets are mined. Then, the KCCA method is used to perform correlation analysis on the comprehensive representative features, thereby mining the common correlation features between high-dimensional multivariate data sets. Finally, due to the analysis results based on KCCA, multiple groups of kernel typical correlation relationships will be generated, in which each pair of correlation relationships has many-to-many correlations, forming complex organizational features. Therefore, the visual analysis technology can be combined to realize the effective extraction of common correlation features between multiple groups of data sets; at the same time, in order to solve the problem that the analysis results are redundant and complex and difficult to intuitively identify, the Sankey diagram can quickly explore the kernel typical correlation patterns between high-dimensional multivariate complex data sets, thereby accurately mining the correlation size and correlation factors. At the same time, by studying the proportion of the total sample variance in the corresponding comprehensive characteristics that can be explained by the indicator characteristics, and visually displaying the proportion of each indicator characteristic in the corresponding comprehensive characteristics (i.e., the corresponding values ​​of UX and VY), the validity of the kernel typical association model can be tested.

[0138] That is to say, in this application, in order to solve the problem of calculation errors caused by inconsistent step sizes due to the temporal nature of the data set, a dynamic time warping algorithm can be used to improve the analysis accuracy. Moreover, the method of this application solves the problem that traditional association analysis technology is difficult to effectively analyze the common correlation features between multiple groups of multivariate data sets, and provides an effective solution for association analysis between high-dimensional, multivariate, and time series data sets. Moreover, in order to solve the problem that complex experimental results of association analysis are difficult to intuitively understand, combined with visual analysis technology, novel views are designed from the visual perception level to accelerate the mining and exploration of experimental results, providing a new idea for the fusion of association analysis and visualization.

[0139] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0140] Based on the same inventive concept, the embodiment of the present application also provides a data processing device for implementing the data processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the one or more data processing device embodiments provided below can refer to the limitations on the data processing method above, and will not be repeated here.

[0141] In one embodiment, Fig.10As shown, a data processing device is provided, including: an acquisition module 1002, a determination module 1004, an analysis module 1006 and a processing module 1008, wherein: the acquisition module 1002 is used to acquire a first time series sequence and a second time series sequence; the first time series sequence includes multiple first data sets; the second time series sequence includes multiple second data sets; the determination module 1004 is used to determine a first target data set from each first data set and a second target data set corresponding to the first target data set in each second data set based on a dynamic regularization path between the first time series sequence and the second time series sequence; the analysis module 1006 is used to perform a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and to perform a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; the processing module 1008 is used to perform a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result between the first time series and the second time series.

[0142] In one embodiment, the processing module 1008 is also used to: obtain a first weight feature that matches the first comprehensive feature, and a second weight feature that matches the second comprehensive feature; based on the first weight feature, the second weight feature and the Gaussian kernel function, perform a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain the correlation analysis results of the first time series sequence and the second time series sequence.

[0143] In one embodiment, the first comprehensive feature includes multiple first sub-features; the processing module 1008 is also used to: obtain multiple first candidate weight features; for each first candidate weight feature, perform weighted summation of the first candidate weight feature to each first sub-feature to determine the first indicator feature corresponding to the first candidate weight feature; the first indicator feature characterizes the degree of aggregation of effective features in the first time series sequence; based on the redundant analysis results of the first comprehensive features matching each first indicator feature, the first candidate weight feature corresponding to the largest redundant analysis result is determined as the first weight feature matching the first comprehensive feature.

[0144] In one embodiment, there are multiple first weight features, each of which includes multiple first sub-weights. The processing module 1008 is also used to: map each first sub-feature and the first indicator feature corresponding to each first weight feature based on the first sub-weight corresponding to each first sub-feature in each first weight feature, determine and use a Sankey diagram to visualize the first mapping result; wherein the first mapping result includes multiple first sub-mapping results, and the first sub-mapping result represents the importance of the first sub-feature in the first comprehensive feature; wherein the first node in the visualized Sankey diagram represents each first sub-feature in the first comprehensive feature, and the second node in the visualized Sankey diagram represents the first indicator feature corresponding to each first weight feature; the edge between the first node and the second node represents each first sub-weight corresponding to the first sub-feature; the indicator value on the first indicator feature is the sum of the first sub-weights corresponding to each mapped first sub-feature.

[0145] In one embodiment, the processing module 1008 is further configured to: determine a redundancy analysis result of the first comprehensive feature matching the first indicator feature based on a ratio between the variance of the first indicator feature and the total variance of the first comprehensive feature.

[0146] In one embodiment, there are multiple first target data sets, and each first target data set includes multiple first target features; the analysis module 1006 is also used to: perform dimensionality reduction processing on each first target feature in each first target data set to obtain the reduced dimensionality feature corresponding to the first target data set; combine the reduced dimensionality features corresponding to each first target data set to obtain the first comprehensive feature corresponding to the first time series.

[0147] Each module in the above data processing device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0148] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.11As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store XX data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a data processing method is implemented.

[0149] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0150] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0151] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0152] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0153] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0154] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0155] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire a first time series sequence and a second time series sequence; the first time series sequence includes a plurality of first data sets; the second time series sequence includes a plurality of second data sets; Determine, based on a dynamically regularized path between the first time series and the second time series, a first target data set from each of the first data sets and a second target data set corresponding to the first target data set from each of the second data sets; Performing a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and performing a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; Performing kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain correlation analysis results of the first time series and the second time series.

2. The method according to claim 1, characterized in that The performing kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain the correlation analysis result of the first time series sequence and the second time series sequence includes: Obtaining a first weighted feature that matches the first comprehensive feature, and a second weighted feature that matches the second comprehensive feature; Based on the first weight feature, the second weight feature and the Gaussian kernel function, a kernel canonical correlation analysis is performed on the first comprehensive feature and the second comprehensive feature to obtain an association analysis result of the first time series sequence and the second time series sequence.

3. The method according to claim 2, characterized in that The first comprehensive feature includes a plurality of first sub-features; and the method further includes: Obtaining multiple first candidate weight features; For each of the first candidate weighted features, weighted summing is performed on each of the first sub-features by using the first candidate weighted feature to determine a first indicator feature corresponding to the first candidate weighted feature; the first indicator feature represents the degree of aggregation of effective features in the first comprehensive feature; Based on the redundant analysis results of the first comprehensive features matching the first indicator features, a first candidate weight feature corresponding to the largest redundant analysis result is determined as a first weight feature matching the first comprehensive feature.

4. The method according to claim 3, characterized in that There are multiple first weight features, each of which includes multiple first sub-weights, and the method further includes: Based on the first sub-weights corresponding to the first sub-features in the first weight features, mapping the first sub-features and the first indicator features corresponding to the first weight features, determining and visually displaying the first mapping result using a Sankey diagram; wherein the first mapping result includes a plurality of first sub-mapping results, each of which represents the importance of the first sub-feature in the first comprehensive feature; Among them, the first node in the visually displayed Sankey diagram represents each of the first sub-features in the first comprehensive feature, and the second node in the visually displayed Sankey diagram represents the first indicator feature corresponding to each of the first weight features; the edge between the first node and the second node represents each of the first sub-weights corresponding to the first sub-feature; the indicator value on the first indicator feature is the sum of the first sub-weights corresponding to each of the mapped first sub-features.

5. The method according to claim 3, characterized in that: The method further comprises: Based on the ratio between the variance of the first indicator feature and the total variance of the first comprehensive feature, a redundancy analysis result of the first comprehensive feature matching the first indicator feature is determined.

6. The method according to claim 1, characterized in that There are multiple first target data sets, and each of the first target data sets includes multiple first target features; The performing dimensionality reduction analysis on the first target data set to determine the first comprehensive feature corresponding to the first time series sequence includes: For each of the first target features in each of the first target data sets, perform dimensionality reduction processing on each of the first target features in the first target data sets to obtain a dimensionality reduction feature corresponding to the first target data set; The dimensionality reduction features corresponding to the first target data sets are combined to obtain a first comprehensive feature corresponding to the first time series.

7. A data processing device, characterized in that: The device comprises: An acquisition module, configured to acquire a first time series sequence and a second time series sequence; the first time series sequence includes a plurality of first data sets; the second time series sequence includes a plurality of second data sets; a determination module, configured to determine a first target data set from each of the first data sets and a second target data set corresponding to the first target data set from each of the second data sets based on a dynamically regularized path between the first time series sequence and the second time series sequence; An analysis module, configured to perform a dimensionality reduction analysis on the first target data set to determine a first comprehensive feature corresponding to the first time series sequence, and to perform a dimensionality reduction analysis on the second target data set to determine a second comprehensive feature corresponding to the second time series sequence; The processing module is used to perform a kernel canonical correlation analysis on the first comprehensive feature and the second comprehensive feature to obtain a correlation analysis result of the first time series sequence and the second time series sequence.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

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 method according to any one of claims 1 to 6 are implemented.

Images