A distributed storage method for financial data based on cloud computing
By building a multi-dimensional coordinate system in a financial database and using a connected graph split clustering algorithm, the clustering results of financial data are optimized, and the problem of failure to fully consider the impact of data point distribution in the existing technology is solved, and more efficient financial data is achieved storage.
Patent Information
- Application Number
- CN202410622211.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-05-20
AI Technical Summary
The existing distributed storage method for financial data fails to fully consider the impact of the distribution of data points in each connected graph on the clustering results when clustering, resulting in data with relatively similar correlations and data with larger differences being clustered into one. class, resulting in low storage efficiency.
By obtaining one-dimensional data sets of multiple dimensions in the financial database, building two-dimensional and four-dimensional coordinate systems, calculating the data trend differences in each dimension, and clustering data points using the connected graph split clustering algorithm, according to each The overall trend difference and split threshold optimization clustering results of individual clusters are ultimately achieved with efficient financial data storage.
By conducting detailed analysis and clustering of financial data from multiple dimensions, the accuracy and practicality of clustering results are improved, making the storage efficiency of financial data higher.
Smart Images

Figure CN118535943B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a distributed storage method for financial data based on cloud computing. Background Art
[0002] In the process of optimizing and analyzing the distributed storage of financial data, data is usually backed up and redundantly stored to prevent data loss, while also supporting data recovery and failover. Because some dimensions of financial data have a strong correlation with other dimensions of financial data, while some dimensions of financial data have no strong correlation with other dimensions of financial data, when clustering financial data, it is necessary to cluster according to the correlation between each dimension of financial data and other dimensions of financial data.
[0003] When clustering all financial data using the connected graph splitting clustering algorithm, only the correlation of the distribution of financial data points in the overall financial data is considered, without considering the impact of the distribution of data points in each connected graph on the clustering results. As a result, data with relatively similar correlations and data with large differences in correlations are clustered into one category, resulting in low efficiency in data storage based on the clustering results of financial data. Summary of the invention
[0004] The present invention provides a financial data distributed storage method based on cloud computing to solve the existing problems.
[0005] A distributed storage method for financial data based on cloud computing of the present invention adopts the following technical solution:
[0006] The present invention proposes a distributed storage method for financial data based on cloud computing, which includes the following steps:
[0007] In a financial database, a one-dimensional data set of multiple dimensions is obtained, wherein each dimension data in the one-dimensional data set corresponds to a relevant date; and multiple data sequences are obtained according to the relevant date corresponding to each dimension data;
[0008] With the relevant dates and values as coordinate axes, a two-dimensional coordinate system is constructed to obtain the corresponding principal component values of each dimension of data; with each dimension as the coordinate axis, a four-dimensional coordinate system is constructed to obtain the data trend difference of each dimension based on the corresponding principal component values of each dimension of data and the distribution of data points in the four-dimensional coordinate system;
[0009] Cluster the data points in the four-dimensional coordinate system to obtain multiple clusters, multiple split-connected graphs, and the length of the broken nodes between each split-connected graph and the adjacent split-connected graph; obtain the overall trend difference of each data point based on the data of each dimension in the data sequence corresponding to each data point and the data trend difference of each dimension; obtain the possibility of each cluster not being optimized based on the overall trend difference of each data point in each cluster and the number of data points contained, and obtain the clusters to be divided, the clusters not to be divided, and the split-connected graphs to be divided;
[0010] According to the possibility of each cluster not being optimized and the length of the broken nodes between each split connected graph and the adjacent split connected graph, the split threshold of each cluster to be divided is obtained; according to the split threshold of each cluster to be divided, the final clustering result is obtained, and the financial data is stored.
[0011] Furthermore, in the financial database, a one-dimensional data set of multiple dimensions is obtained, wherein each dimension data in the one-dimensional data set corresponds to a relevant date; and multiple data sequences are obtained according to the relevant date corresponding to each dimension data, including the specific method of:
[0012] In the financial database, data on four dimensions, namely monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate, are collected for each month in M years to obtain a one-dimensional data set of four dimensions, where M is the preset collection year;
[0013] The dimensional data with the same relevant dates corresponding to different dimensions are recorded as a data series to obtain multiple data series.
[0014] Furthermore, the two-dimensional coordinate system is constructed with the relevant dates and values as coordinate axes to obtain the corresponding principal component values of each dimensional data, including the specific method of:
[0015] Map all the data on the monthly profit growth rate dimension to a two-dimensional coordinate system with the horizontal axis being the relevant date and the vertical axis being the data value;
[0016] Through principal component analysis, we obtain the projection value of each month's profit growth rate data in the direction of the largest principal component in the two-dimensional coordinate system of date-value, and record it as the corresponding principal component value of each month's profit growth rate data.
[0017] Furthermore, the four-dimensional coordinate system is constructed with each dimension as the coordinate axis, and the data trend difference of each dimension is obtained according to the corresponding principal component value of each dimension data and the distribution of data points in the four-dimensional coordinate system, including the specific method of:
[0018] A four-dimensional coordinate system is constructed with the four dimensions of monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate as coordinate axes;
[0019] The corresponding principal component value of any dimension data in any data sequence is used as the coordinate of the data sequence on the corresponding coordinate axis of the dimension, and all data sequences are mapped into the four-dimensional coordinate system, where the j-th data sequence corresponds to the j-th data point in the four-dimensional coordinate system, and the j-th data sequence is composed of the j-th dimension data on four dimensions;
[0020] Calculate the Euclidean distance between each data point and other data points in the four-dimensional coordinate system, and record the k data points with the smallest Euclidean distance to any data point as the neighboring data points of the data point, where k is the preset number of neighboring data points;
[0021] According to the Euclidean distance between each data point and its neighboring data points in the four-dimensional coordinate system, and the corresponding principal component value of each data point corresponding to each dimension of data, the data trend difference of each dimension is obtained.
[0022] Furthermore, the data trend difference of each dimension is obtained according to the Euclidean distance between each data point and the adjacent data points in the four-dimensional coordinate system, and the corresponding principal component value of each data point corresponding to each dimensional data, including the specific method of:
[0023]
[0024] In the formula, δ i represents the data trend difference of the i-th dimension, Q represents the number of data contained in each one-dimensional data set, k represents the number of neighboring data points of each data point, and d i,j,c It represents the Euclidean distance between the jth data point and the cth neighboring data point in the four-dimensional coordinate system, l i,j It represents the corresponding principal component value of the j-th dimension data in the i-th dimension one-dimensional data set, and norm() represents the normalization function.
[0025] Furthermore, the data points in the four-dimensional coordinate system are clustered to obtain multiple clusters, multiple split-connected graphs, and the length of the broken nodes between each split-connected graph and the adjacent split-connected graph, including the specific method of:
[0026] Using Euclidean distance as the distance metric, a connectivity graph is constructed for the data points in the four-dimensional coordinate system through a connectivity graph splitting clustering algorithm to obtain a constructed connectivity graph, and then several split connectivity graphs are obtained; each split connectivity graph corresponds to a cluster;
[0027] The length of the node connecting each split connected graph with other split connected graphs in the constructed connected graph is recorded as the broken node length, and the broken node length set of each split connected graph is obtained; the number of occurrences of the value of the broken node length in the broken node length set of any split connected graph is recorded as the number of the node lengths connecting the split connected graph with other split connected graphs in the constructed connected graph;
[0028] In the constructed connected graph, there are split connected graphs with nodes connected to each other, which are adjacent split connected graphs to each other, and the adjacent split connected graphs of each split connected graph are obtained.
[0029] Furthermore, the overall trend difference of each data point is obtained according to the data of each dimension in the data sequence corresponding to each data point and the data trend difference of each dimension, including the specific method of:
[0030]
[0031] In the formula, δ a,b represents the overall trend difference of the bth data point in the ath cluster, m represents the number of dimensions, and δ d Indicates the data trend difference of the dth dimension, l a,b,d Indicates the corresponding principal component value of the b-th data point in the a-th cluster corresponding to the d-th dimension data.
[0032] Furthermore, the possibility of not optimizing each cluster is obtained according to the overall trend difference of each data point in each cluster and the number of data points contained therein, and the split connectivity graph of the clusters to be divided, the clusters not to be divided, and the clusters to be divided is obtained, including the specific method of:
[0033]
[0034] In the formula, C a represents the possibility of not optimizing the a-th cluster, δ a,b Indicates the overall trend difference of the bth data point in the ath cluster, Represents the mean of the overall trend difference of all data points in the a-th cluster, Indicates the maximum absolute value of the difference between the overall trend difference of each data point in the a-th cluster and the mean of the overall trend difference of all data points, N a represents the number of data points contained in the a-th cluster, norm() represents the normalization function, || represents the absolute value function, and exp() represents the exponential function with a natural constant as the base;
[0035] A possibility threshold T5 is preset, and clusters whose probability of not being optimized is less than T5 are recorded as clusters to be divided, and clusters whose probability of not being optimized is greater than or equal to T5 are recorded as clusters not to be divided;
[0036] The split connected graph corresponding to the cluster to be divided is recorded as the split connected graph to be divided.
[0037] Furthermore, the splitting threshold of each cluster to be divided is obtained according to the possibility of each cluster not being optimized and the length of the broken node between each split connected graph and the adjacent split connected graph, including the specific method of:
[0038]
[0039] In the formula, LF w,r Indicates the length of the fracture node L 1,w,r is the credibility of the splitting threshold of the w-th cluster to be divided, L 1,w,r represents the length of the rth broken node in the broken node length set of the wth split connected graph to be divided, N w,r Indicates the length of the fracture node L 1,w,r To construct the number of node lengths of the wth split connected graph to be divided in the connected graph and the adjacent split connected graph, CE w Indicates the possibility of not optimizing the wth cluster to be divided, CE w,r,y Indicates that the nodes in the constructed connected graph are connected to the wth split connected graph to be divided and the length of the connected nodes is L 1,w,r The possibility that the clusters corresponding to the split connected graph are not optimized;
[0040] The length of all broken nodes of any cluster to be divided that has the largest credibility of the split threshold of the cluster to be divided is used as the split threshold of the split connected graph to be divided.
[0041] Furthermore, the final clustering result is obtained according to the splitting threshold of each cluster to be divided, including the specific method of:
[0042] Disconnect the line segments between nodes in any split connected graph to be divided whose length is greater than the split threshold of the split connected graph to be divided, to obtain a new split connected graph to be divided;
[0043] Continue to re-partition the new split-connected graph to be divided through the connected graph split clustering algorithm to obtain the final split-connected graph;
[0044] New clusters are obtained through the final split connectivity graph; the new clusters and the undivided clusters are taken as the final clustering results.
[0045] The beneficial effects of the technical solution of the present invention are as follows: the present invention analyzes financial data of multiple dimensions, constructs a coordinate system through multiple dimensions, and calculates the data trend difference of each dimension based not only on the corresponding principal component values of all data in the one-dimensional data set of each dimension, but also on the distribution of data points, laying the foundation for obtaining the final clustering result; when calculating the overall trend difference of each data point, it is based not only on the corresponding principal component values of each dimensional data in the data sequence corresponding to each data point, but also on the data trend difference of each dimension, so that the clusters corresponding to the split connected graph whose internal data point distribution has little effect on the clustering result are more likely not to be optimized; when obtaining the split nodes of each cluster to be divided, based on the possibility that each cluster will not be optimized, the split threshold of each cluster to be divided is made more accurate, resulting in a more practical final clustering result and higher storage efficiency of financial data. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0047] Figure 1 This is a flowchart of the steps of a distributed storage method for financial data based on cloud computing of the present invention;
[0048] Figure 2 is a schematic diagram of the length of the fracture node;
[0049] Figure 3 Schematic diagram of the adjacent split connectivity graph. DETAILED DESCRIPTION
[0050] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a distributed storage method for financial data based on cloud computing proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0051] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0052] The following is a detailed description of a specific scheme of a distributed storage method for financial data based on cloud computing provided by the present invention in conjunction with the accompanying drawings.
[0053] See also Figure 1 , which shows a flowchart of a method for distributed storage of financial data based on cloud computing provided by an embodiment of the present invention, the method comprising the following steps:
[0054] Step S001: In a financial database, obtain a one-dimensional data set of multiple dimensions; and obtain multiple data sequences according to the relevant date corresponding to each dimensional data.
[0055] The purpose of this embodiment is to optimize the distributed storage method for financial data, thereby obtaining financial data in multiple dimensions.
[0056] Specifically, in a financial database, data on four dimensions, namely, monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate, are collected for each month in M years to obtain a one-dimensional data set of four dimensions. Each dimension data in each one-dimensional data set corresponds to a relevant date. The preset collection year M=10 in this embodiment is used as an example for description, and other implementations may be set to other values. The dimension data corresponding to the same relevant date on different dimensions are recorded as a data sequence to obtain multiple data sequences. The j-th dimension data in the one-dimensional data set on four dimensions constitutes the j-th data sequence.
[0057] Step S002: Construct a two-dimensional coordinate system with relevant dates and values as coordinate axes to obtain the corresponding principal component values of each dimension data; construct a four-dimensional coordinate system with each dimension as the coordinate axis, and obtain the data trend difference of each dimension based on the corresponding principal component values of each dimension data and the distribution of data points in the four-dimensional coordinate system.
[0058] It should be noted that, because this embodiment clusters data based on the correlation between financial data, the correlation between data of different dimensions is calculated. Each dimensional data has a time series variation feature, so the correlation between data of different dimensions is quantified according to the similarity between the time series variation features of different dimensional data and the time series variation features of other dimensional data. Therefore, the time series variation features of each dimensional data, that is, the corresponding principal component value of each dimensional data, are calculated by principal component analysis. The time series variation features of each dimensional data in each data sequence are obtained.
[0059] It should be further explained that the similarity of the corresponding principal component values of a data series and other data series in multiple dimensions of the same data reflects the degree of correlation between the overall dimensions. Therefore, a four-dimensional coordinate system is constructed with each dimension as the coordinate axis. According to the distribution of data points in the four-dimensional coordinate system, the difference in data trend in each dimension is obtained.
[0060] Specifically, among the four dimensions, take the data on the monthly profit growth rate dimension as an example. All data on the monthly profit growth rate dimension are mapped to a two-dimensional coordinate system with the horizontal axis being the relevant date and the vertical axis being the data value. Through the principal component analysis method, the projection value of each monthly profit growth rate data in the direction of the largest principal component in the two-dimensional coordinate system of date-value is obtained. The projection value of the monthly profit rate growth data in the direction of the largest principal component in the two-dimensional coordinate system of date-value is recorded as the corresponding principal component value of the monthly profit growth rate data. Among them, the principal component analysis method is a well-known technology and will not be repeated in this embodiment.
[0061] Furthermore, through the above method, the corresponding principal component value of each dimension data in the three dimensions of monthly turnover, monthly turnover rate, and monthly financial return rate is obtained, and then the corresponding principal component value of each dimension data in each data sequence is obtained.
[0062] Furthermore, a four-dimensional coordinate system is constructed with the four dimensions of monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate as coordinate axes. All data sequences are mapped into the four-dimensional coordinate system. The corresponding principal component value of any dimension data in any data sequence is the coordinate of the data sequence on the corresponding coordinate axis of the dimension. The jth data point in the four-dimensional coordinate system corresponds to the jth data sequence, that is, the jth data point in the four-dimensional coordinate system corresponds to the jth dimension data on the four dimensions.
[0063] Furthermore, the Euclidean distance between each data point and other data points in the four-dimensional coordinate system is calculated, and the k data points with the smallest Euclidean distance to any data point are recorded as the neighboring data points of the data point. In this embodiment, the number of neighboring data points k=15 is preset, and this is used as an example for description. Other values can be set in other implementations.
[0064] Furthermore, according to the distribution of data points in the four-dimensional coordinate system, the data trend difference of each dimension is obtained, and the data trend difference of the i-th dimension is recorded as δ i , the specific calculation formula is as follows:
[0065]
[0066] In the formula, δ i represents the data trend difference of the i-th dimension, Q represents the number of data contained in each one-dimensional data set, k represents the number of neighboring data points of each data point, and d i,j,c It represents the Euclidean distance between the jth data point and the cth neighboring data point in the four-dimensional coordinate system, l i,jRepresents the corresponding principal component value of the j-th dimension data in the one-dimensional data set of the i-th dimension, norm() represents the linear normalization function, and its normalized object is each dimension
[0067] It should be noted that It represents the difference between the significant status of the dimensional change trend of the j-th dimension data in the ith dimension one-dimensional data set during the dimensionality reduction process and the significant status of the distribution of the corresponding data points in the four-dimensional coordinate system. The larger the value, the smaller the correlation between the data of the i-th dimension and the data of other dimensions. It further indicates that the smaller the correlation between the data of the i-th dimension and the data of other dimensions, that is, the greater the difference in the data trend of the i-th dimension.
[0068] It should be further explained that δ i It represents the significance of the data of the i-th dimension in the distribution relationship of the overall multidimensional data. The larger its value, the weaker the correlation between the data of the i-th dimension and the overall multidimensional data. For any two dimensions, the absolute value of the difference between the data trend difference of each dimension and the data trend difference of the remaining dimensions indicates the correlation between the data of the two dimensions in time series changes.
[0069] At this point, the data trend differences in each dimension are obtained.
[0070] Step S003: Cluster the data points in the four-dimensional coordinate system to obtain multiple clusters, multiple split-connected graphs, and the length of the broken nodes between each split-connected graph and the adjacent split-connected graph; obtain the overall trend difference of each data point based on the data of each dimension in the data sequence corresponding to each data point and the data trend difference of each dimension; obtain the possibility of each cluster not being optimized based on the overall trend difference of each data point in each cluster and the number of data points contained, and obtain the clusters to be divided, the clusters not to be divided, and the split-connected graphs to be divided.
[0071] It should be noted that, since the main purpose of this embodiment is to post-process the clustering results, the data points in the four-dimensional coordinate system need to be clustered to obtain multiple clusters.
[0072] Specifically, using Euclidean distance as the distance metric, a connectivity graph is constructed for the data points in the four-dimensional coordinate system through a connectivity graph splitting clustering algorithm to obtain a constructed connectivity graph; further, a number of split connectivity graphs are obtained. Each split connectivity graph corresponds to a cluster, that is, the data point in the a-th split connectivity graph is the data point in the a-th cluster. In addition, the parameters contained in the connectivity graph splitting clustering algorithm are processed with the default parameters of the algorithm itself.
[0073] Furthermore, the length of the nodes in each split connected graph connected to other split connected graphs in the constructed connected graph is recorded as the broken node length, where the broken node length is as follows: Figure 2 As shown. The length set of the broken nodes of each split connected graph is obtained. The length set of the broken nodes of the a-th split connected graph is L a ={…,L a,f ,…}, L a,f It represents the length of the fth broken node of the ath split-connected graph. The number of times the value of the broken node length in the broken node length set of any split-connected graph appears is recorded as the number of node lengths that the split-connected graph is connected to other split-connected graphs in the constructed connected graph. In addition, there is no identical value in the broken node length set of each split-connected graph.
[0074] Furthermore, in the constructed connectivity graph, there are split connectivity graphs with nodes connected to each other, which are adjacent split connectivity graphs, and the adjacent split connectivity graphs of each split connectivity graph are obtained. The distribution relationship between each split connectivity graph and its adjacent split connectivity graph is as follows: Figure 3 shown.
[0075] It should be noted that when clustering data points in the four-dimensional coordinate system through the connected graph splitting clustering algorithm, only the overall distribution relationship of the data points in the four-dimensional coordinate system is considered, and the distribution relationship of the data points around each data point is not considered. That is, the distribution relationship of the data points in each connected graph affects the clustering results. As a result, in some clusters, the Euclidean distance between most data points is small, while the Euclidean distance between a small number of data points in such clusters and the remaining data points is large; that is, in some clusters, the distribution of some data points and the surrounding data points is quite different from the overall distribution of the data points in the cluster where this data point is located. Because the Euclidean distance between each data point and the remaining data points is determined by the corresponding principal component value of each dimensional data in the data sequence corresponding to each data point, the possibility of not optimizing each cluster is obtained according to the corresponding principal component value of each dimensional data in the data sequence corresponding to each data point and the difference in the data trend of each dimension.
[0076] Specifically, according to the principal component value of each dimension data in the data sequence corresponding to each data point and the difference in the data trend of each dimension, the possibility of not optimizing each cluster is obtained. The possibility of not optimizing the a-th cluster is recorded as C a , the specific calculation formula is as follows:
[0077]
[0078]
[0079] In the formula, δ a,brepresents the overall trend difference of the bth data point in the ath cluster, m represents the number of dimensions, and δ d Indicates the data trend difference of the dth dimension, l a,b,d Indicates the principal component value of the bth data point in the ath cluster corresponding to the dth dimension data, C a represents the possibility of not optimizing the a-th cluster. represents the mean of the overall trend difference of all data points in the a-th cluster, || represents the absolute value function, Indicates the maximum absolute value of the difference between the overall trend difference of each data point in the a-th cluster and the mean of the overall trend difference of all data points, N a represents the number of data points contained in the a-th cluster, exp() is an exponential function with a natural constant as the base, and this embodiment adopts the exp(-x) model to present the inverse proportional relationship and normalization processing, x is the input of the model, and the implementer can set the inverse proportional function and normalization function according to the actual situation; norm() represents a linear normalization function, and its normalization object is the
[0080]
[0081] It should be noted that the possibility of not optimizing the a-th cluster represents the correlation between the overall trend difference of each data point in the a-th cluster and the actual position distribution. If the overall trend difference of the data points in the cluster is greater, it proves that there is a large difference in the financial data in this cluster, and there is no clustering of position distribution in the actual sample space, then the division result of the cluster is considered to be more credible, that is, the possibility that this cluster will not be divided is greater.
[0082] Furthermore, a possibility threshold T5 is preset, and clusters whose possibility of not being optimized is less than T5 are recorded as clusters to be divided, and all clusters to be divided are obtained. The split connected graph corresponding to the cluster to be divided is recorded as the split connected graph to be divided. Clusters whose possibility of not being optimized is greater than or equal to T5 are recorded as clusters not to be divided.
[0083] The preset possibility threshold value T5=0.38 in this embodiment is used as an example for description. Other values may be set in other implementation modes.
[0084] At this point, the possibility of not optimizing each cluster, the clusters to be divided, or the split connectivity graph are obtained.
[0085] Step S004: According to the possibility of each cluster not being optimized and the length of the broken nodes between each split connected graph and the adjacent split connected graph, the split threshold of each cluster to be divided is obtained; according to the split threshold of each cluster to be divided, the final clustering result is obtained, and the financial data is stored.
[0086] It should be noted that the possibility of each cluster not being optimized reflects the credibility of the length of the broken nodes in the split connected graph corresponding to the cluster as the split threshold. Therefore, according to the possibility of each cluster not being optimized, the split threshold of each cluster or decomposed connected graph is obtained.
[0087] Specifically, according to the possibility of not optimizing each cluster, the length of the broken node L is obtained. 1,w,r is the credibility of the splitting threshold of the w-th cluster to be divided. The specific calculation formula is as follows:
[0088]
[0089] In the formula, LF w,r Indicates the length of the fracture node L 1,w,r is the credibility of the splitting threshold of the w-th cluster to be divided, L 1,w,r represents the length of the rth broken node in the broken node length set of the wth split connected graph to be divided, N w,r Indicates the length of the fracture node L 1,w,r To construct the number of node lengths of the wth split connected graph to be divided in the connected graph and the adjacent split connected graph, CE w Indicates the possibility of not optimizing the wth cluster to be divided, CE w,r,y Indicates that the nodes in the constructed connected graph are connected to the wth split connected graph to be divided and the length of the connected nodes is L 1,w,r The possibility that the clusters corresponding to the split connected graph are not optimized.
[0090] Furthermore, the credibility of the splitting threshold of the cluster to be divided is obtained for the length of each broken node of each cluster to be divided, and the length of all broken nodes of any cluster to be divided that has the largest credibility of the splitting threshold of the cluster to be divided is used as the splitting threshold of the split connected graph to be divided.
[0091] Furthermore, the line segments between nodes in any split-connected graph to be divided whose length is greater than the split threshold of the split-connected graph are disconnected to obtain a new split-connected graph to be divided. The new split-connected graph to be divided is re-divided through the previous connected graph split clustering algorithm to obtain the final split-connected graph. New clusters are obtained through the final split-connected graph. The new clusters and the undivided clusters are taken as the final clustering results.
[0092] Furthermore, based on the final clustering result, the financial data is distributedly stored, wherein the financial data is classified according to the final clustering result, thereby realizing distributed storage of the financial data. The existing method is used for distributed storage of the financial data, which will not be described in detail in this embodiment.
[0093] At this point, this embodiment is completed.
[0094] To sum up, in the embodiments of the present invention, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A distributed storage method for financial data based on cloud computing. It is characterized in that The method comprises the following steps: In a financial database, a one-dimensional data set of multiple dimensions is obtained, wherein each dimension data in the one-dimensional data set corresponds to a relevant date; and multiple data sequences are obtained according to the relevant date corresponding to each dimension data; With the relevant dates and values as coordinate axes, a two-dimensional coordinate system is constructed to obtain the corresponding principal component values of each dimension of data; with each dimension as the coordinate axis, a four-dimensional coordinate system is constructed to obtain the data trend difference of each dimension based on the corresponding principal component values of each dimension of data and the distribution of data points in the four-dimensional coordinate system; Cluster the data points in the four-dimensional coordinate system to obtain multiple clusters, multiple split-connected graphs, and the length of the broken nodes between each split-connected graph and the adjacent split-connected graph; obtain the overall trend difference of each data point based on the data of each dimension in the data sequence corresponding to each data point and the data trend difference of each dimension; obtain the possibility of each cluster not being optimized based on the overall trend difference of each data point in each cluster and the number of data points contained, and obtain the clusters to be divided, the clusters not to be divided, and the split-connected graphs to be divided; According to the possibility of each cluster not being optimized and the length of the broken nodes between each split connected graph and the adjacent split connected graph, the split threshold of each cluster to be divided is obtained; according to the split threshold of each cluster to be divided, the final clustering result is obtained, and the financial data is stored; The method of clustering the data points in the four-dimensional coordinate system to obtain multiple clusters, multiple split-connected graphs, and the length of the broken nodes between each split-connected graph and the adjacent split-connected graph includes: Using Euclidean distance as the distance metric, a connectivity graph is constructed for the data points in the four-dimensional coordinate system through a connectivity graph splitting clustering algorithm to obtain a constructed connectivity graph, and then several split connectivity graphs are obtained; each split connectivity graph corresponds to a cluster; The length of the node connecting each split connected graph with other split connected graphs in the constructed connected graph is recorded as the broken node length, and the broken node length set of each split connected graph is obtained; the number of occurrences of the value of the broken node length in the broken node length set of any split connected graph is recorded as the number of the node lengths connecting the split connected graph with other split connected graphs in the constructed connected graph; In the constructed connected graph, there are split connected graphs with nodes connected to each other, which are adjacent split connected graphs to each other, and the adjacent split connected graph of each split connected graph is obtained; The specific method of obtaining the overall trend difference of each data point according to the data of each dimension in the data sequence corresponding to each data point and the data trend difference of each dimension is as follows: In the formula, δ a,b represents the overall trend difference of the bth data point in the ath cluster, m represents the number of dimensions, and δ d Indicates the data trend difference of the dth dimension, l a,b,d Indicates the corresponding principal component value of the d-th dimension data corresponding to the b-th data point in the a-th cluster; The method of obtaining the possibility of not optimizing each cluster according to the overall trend difference of each data point in each cluster and the number of data points contained therein, and obtaining the split connectivity graph of the clusters to be divided, the clusters not to be divided, and the clusters to be divided includes the following specific methods: In the formula, C a represents the possibility of not optimizing the a-th cluster, δ a,b Indicates the overall trend difference of the bth data point in the ath cluster, Represents the mean of the overall trend difference of all data points in the a-th cluster, Indicates the maximum absolute value of the difference between the overall trend difference of each data point in the a-th cluster and the mean of the overall trend difference of all data points, N a represents the number of data points contained in the a-th cluster, norm() represents the normalization function, || represents the absolute value function, and exp() represents the exponential function with a natural constant as the base; Preset probability threshold T 5 , the probability of not optimizing is less than T 5 The clusters are recorded as clusters to be divided, and the probability of not optimizing is greater than or equal to T 5 The clusters of are recorded as unclassified clusters; The split connected graph corresponding to the cluster to be divided is recorded as the split connected graph to be divided; The method of obtaining the split threshold of each cluster to be divided according to the possibility of not optimizing each cluster and the length of the broken node between each split connected graph and the adjacent split connected graph includes the following specific methods: In the formula, LF w,r Indicates the length of the fracture node L 1,w,r is the credibility of the splitting threshold of the w-th cluster to be divided, L 1,w,r represents the length of the rth broken node in the broken node length set of the wth split connected graph to be divided, N w,r Indicates the length of the fracture node L 1,w,r To construct the number of node lengths of the wth split connected graph to be divided in the connected graph and the adjacent split connected graph, CE w Indicates the possibility of not optimizing the wth cluster to be divided, CE w,r,y Indicates that the nodes in the constructed connected graph are connected to the wth split connected graph to be divided and the length of the connected nodes is L 1,w,r The possibility that the clusters corresponding to the split connected graph are not optimized; The length of all broken nodes of any cluster to be divided that has the largest credibility of the split threshold of the cluster to be divided is used as the split threshold of the split connected graph to be divided.
2. According to the cloud computing-based distributed storage method for financial data according to claim 1, It is characterized in that The method of obtaining a one-dimensional data set of multiple dimensions in a financial database, wherein each dimension data in the one-dimensional data set corresponds to a relevant date, and obtaining multiple data sequences according to the relevant date corresponding to each dimension data, includes the following specific methods: In the financial database, data on four dimensions, namely monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate, are collected for each month in M years to obtain a one-dimensional data set of four dimensions, where M is the preset collection year; The dimensional data with the same relevant dates corresponding to different dimensions are recorded as a data series to obtain multiple data series.
3. According to the cloud computing-based distributed storage method for financial data according to claim 2, It is characterized in that The method of constructing a two-dimensional coordinate system with relevant dates and values as coordinate axes to obtain the corresponding principal component values of each dimension data includes the following specific methods: Map all the data on the monthly profit growth rate dimension to a two-dimensional coordinate system with the horizontal axis being the relevant date and the vertical axis being the data value; Through principal component analysis, we obtain the projection value of each month's profit growth rate data in the direction of the largest principal component in the two-dimensional coordinate system of date-value, and record it as the corresponding principal component value of each month's profit growth rate data.
4. According to claim 3, a distributed storage method for financial data based on cloud computing, It is characterized in that The method of constructing a four-dimensional coordinate system with each dimension as the coordinate axis and obtaining the data trend difference of each dimension according to the corresponding principal component value of each dimension data and the distribution of data points in the four-dimensional coordinate system includes the following specific methods: A four-dimensional coordinate system is constructed with the four dimensions of monthly turnover, monthly turnover rate, monthly financial return rate, and monthly profit growth rate as coordinate axes; The corresponding principal component value of any dimension data in any data sequence is used as the coordinate of the data sequence on the corresponding coordinate axis of the dimension, and all data sequences are mapped into the four-dimensional coordinate system, where the j-th data sequence corresponds to the j-th data point in the four-dimensional coordinate system, and the j-th data sequence is composed of the j-th dimension data on four dimensions; Calculate the Euclidean distance between each data point and other data points in the four-dimensional coordinate system, and record the k data points with the smallest Euclidean distance to any data point as the neighboring data points of the data point, where k is the preset number of neighboring data points; According to the Euclidean distance between each data point and its neighboring data points in the four-dimensional coordinate system, and the corresponding principal component value of each data point corresponding to each dimension of data, the data trend difference of each dimension is obtained.
5. According to claim 4, a distributed storage method for financial data based on cloud computing, It is characterized in that The specific method of obtaining the data trend difference of each dimension according to the Euclidean distance between each data point and the adjacent data points in the four-dimensional coordinate system and the corresponding principal component value of each data point corresponding to each dimensional data is as follows: In the formula, δ i represents the data trend difference of the i-th dimension, Q represents the number of data contained in each one-dimensional data set, k represents the number of neighboring data points of each data point, and d i,j,c It represents the Euclidean distance between the jth data point and the cth neighboring data point in the four-dimensional coordinate system, l i,j It represents the corresponding principal component value of the j-th dimension data in the i-th dimension one-dimensional data set, and norm() represents the normalization function.
6. According to claim 1, a distributed storage method for financial data based on cloud computing, It is characterized in that The final clustering result is obtained according to the splitting threshold of each cluster to be divided, including the specific method as follows: Disconnect the line segments between nodes in any split connected graph to be divided whose length is greater than the split threshold of the split connected graph to be divided, to obtain a new split connected graph to be divided; Continue to re-partition the new split-connected graph to be divided through the connected graph split clustering algorithm to obtain the final split-connected graph; New clusters are obtained through the final split connectivity graph; the new clusters and the undivided clusters are taken as the final clustering results.
