CN117931954A - A database data interface management method based on flexible configuration - Google Patents

Abstract

The present invention discloses a database data interface management method based on flexible configuration, comprising: step 1) obtaining data of different attributes in the database; step 2) obtaining the representativeness and benchmark degree of different attributes, obtaining representative attributes and benchmark attributes; in different iterations, constructing attribute bipartite graphs, obtaining the final attribute combination; step 3) performing data sharding and clustering according to the final attribute combination to achieve accurate storage. Beneficial effect: The method performs accurate sharding and clustering by obtaining the best attribute combination to improve query efficiency.

Description

A database data interface management method based on flexible configuration

Technical Field

The present invention relates to the technical field of database data processing, and in particular to a database data interface management method based on flexible configuration.

Background technique

In the simulation calculation call and indicator data application display scenarios, the provision of basic data is extremely important. With the continuous deepening of simulation logic and indicator dimensions, the continuous increase in business needs, the increasing number of access conditions and query results, and the increasing number of code development interfaces, the work of code development interfaces is becoming more and more arduous, which will not only increase the workload, but also seriously affect the customer experience, resulting in a large amount of business loss. The traditional mode of developing data source interfaces is not only labor-intensive and difficult to maintain, but also lacks a unified management, monitoring and abnormal rapid recovery mechanism. Therefore, in order to reduce the workload of demand analysis, design, development, testing and deployment, the interface development efficiency is greatly improved through configurable methods, low-code development is achieved, and database interface resources are managed in a unified manner.

In the scenario of large-scale concurrent queries based on a flexible database data interface management system, such a system may encounter some performance bottlenecks. For example, if many queries involve the same large table, then this table may become a performance bottleneck, resulting in a slow query speed. To solve this problem, the traditional method is to use data sharding and clustering technology. This technology divides the data in the database into several smaller subsets, called "shards". Each shard can be queried and updated separately, thereby improving the parallel processing capabilities of the system. At the same time, by clustering the data in the database and dividing it into a "shard", the data locality can be further improved, thereby improving the query efficiency. However, in the process of data sharding and clustering, if different attribute data are clustered separately, and then different data shards are obtained, a large amount of spatial redundancy will be caused, and the query efficiency will also be reduced.

Summary of the invention

The present invention aims to overcome the deficiencies of the prior art and provides a database data interface management method based on flexible configuration, which is specifically implemented by the following technical solutions:

The database data interface management method based on flexible configuration includes:

Step 1) Obtain data of different attributes in the database;

Step 2) Obtain the representativeness and benchmark degree of different attributes, obtain representative attributes and benchmark attributes; construct attribute bipartite graphs in different iterations to obtain the final attribute combination;

Step 3) Data is sharded and clustered according to the final attribute combination to achieve accurate storage.

A further design of the database data interface management method based on flexible configuration is that in the process of obtaining the attribute combination method in step 2), K-M matching is performed on the bipartite graph, the corresponding K-M matching result is used as the attribute combination result, and the attribute combination result in each iteration process is obtained through iteration, so as to comprehensively obtain the final attribute combination method.

The further design of the database data interface management method based on flexible configuration is that in the step 2), the representativeness of the j-th attribute α _j is obtained according to formula (1),

In formula (1), _Nj represents the number of data types of the jth attribute, max(N) represents the maximum value of the number of data types of all attributes, _Js ′ represents the number of all attributes except the jth attribute; _cvs ( _nj ) represents the coefficient of variation of the data values of the sth attribute other than the _jth attribute when the data value of the jth attribute is the njth data value;

A further design of the database data interface management method based on flexible configuration is that in step 2), the representativeness of all attributes is linearly normalized, and the attribute corresponding to the largest benchmark degree is selected and recorded as the benchmark attribute.

The further design of the database data interface management method based on flexible configuration is that the benchmark degree in step 2) is determined by the overall distribution under the current attribute, specifically: when the representative attributes are the same, the truncated data is obtained, and so on, the truncated data under other representative attributes are obtained. Among all the truncated data, the intersection of the truncated data under all the representative attributes is obtained, and the intersection is the first truncated data. In each truncated data, the benchmark degree γ _j of the j-th attribute is calculated according to formula (2),

In formula (2), _Js ^′ represents the number of all attributes other than the jth attribute; M represents the number of all truncated data; _Rm (s) represents the Pearson correlation coefficient between the data in the mth truncated data under the sth attribute of other than the jth attribute and the data in the mth truncated data under the jth attribute.

A further design of the database data interface management method based on flexible configuration is that the process of obtaining the first truncated segment data is: all the data in the database are arranged, the arranged position sequence number is used as the positioning, the data with the same representative attributes are segmented, and the truncated data under the current tabular attribute is obtained.

The further design of the database data interface management method based on flexible configuration is that the process of obtaining the final attribute combination in step 2) is specifically as follows:

Based on the representative attributes and benchmark attributes obtained, the attribute combination is determined by constructing an attribute bipartite graph, wherein the bipartite graph is a bipartite graph when the representative attributes are set to the same value, wherein the left node of the bipartite graph is set to any non-representative attribute, and the right node is set to other non-representative attributes, and the same attributes in the edges connected between the left and right nodes are not connected, then the corresponding edge weights between the acquired nodes can be used to obtain the attribute bipartite graph;

The matching is performed in an iterative manner. The iterative process is as follows: in the first iteration, the K-M matching of the bipartite graph is performed. After the attribute combination of the matching result is obtained, the difference between the data clustered according to the attribute combination and the data clustered before is calculated. If the difference is less than the preset difference threshold, the iteration is stopped. On the basis of the representative attributes obtained, the benchmark attributes are used as the overall benchmark to calculate the data distribution curves of the attributes represented by the left node and the right node when the representative attributes are the same.

Select the d-th data value of the representative attribute, whose left node is the v-th non-representative attribute node, the data distribution curve of the left node has the benchmark attribute on the horizontal axis, and the v-th non-representative attribute on the vertical axis, which is recorded as the first data distribution curve; the right node is the u-th non-representative attribute node, the data distribution of the right node has the benchmark attribute on the horizontal axis, and the u-th non-representative attribute on the vertical axis, which is recorded as the second data distribution curve, and the difference between the data point on each data distribution curve and the corresponding attribute regular distribution curve is used, and then the similarity of the difference is calculated as the edge weight of the edge weight.

The further design of the database data interface management method based on flexible configuration is that the acquisition process of the regular distribution curve is specifically as follows: the horizontal coordinate of the regular distribution curve is set as the reference attribute, the vertical coordinate is set as the vth non-representative attribute, and each existing horizontal coordinate point is calculated to determine the corresponding vertical coordinate data value, that is, the vertical coordinate value _yq corresponding to the qth horizontal coordinate value is calculated according to formula (3):

In formula (3), H represents the number of clusters after DBSCAN clustering based on the qth benchmark attribute; W _h represents the number of data types in the hth cluster; P _w represents the frequency value of the wth data type; G _w represents the data value of the wth data type; Indicates the distribution characteristics of the cluster where the wth data type in the hth cluster is located, which is represented by the density between the data in the cluster; the vertical coordinate value is obtained by weighted average, through/> Indicates the weight of the data type;

According to the ordinate value corresponding to the acquired horizontal coordinate, a continuous distribution curve is obtained, and the distribution curve is fitted to obtain the corresponding regular distribution curve. The regular distribution curve of the vth non-representative attribute node is recorded as the first regular distribution curve, and the regular distribution curve of the uth non-representative attribute node is recorded as the second regular distribution curve; the ordinate value of the horizontal coordinate of the data point on the first data distribution curve is subtracted from the ordinate value of the horizontal coordinate of the first regular distribution curve, and the absolute value of the difference is obtained to obtain the difference value of this horizontal coordinate of the left node, which is recorded as the first difference value. Similarly, the difference value of the same horizontal coordinate of the right node is obtained, which is recorded as is the second difference value, the absolute value of the result obtained by subtracting the ratio of the first difference value and the second difference value of the same horizontal coordinate from 1 is obtained as the difference value of the current horizontal coordinate, and the corresponding average of the difference values of multiple horizontal coordinates is recorded as the difference between the left node and the right node, and the difference is normalized by an inverse proportional function, and the processing result is recorded as the edge weight of the two nodes, and the above operation is repeated to obtain the edge weight between other nodes; K-M matching is performed on the bipartite graph, and the two nodes corresponding to the largest edge weight in the matching result are obtained, recorded as the nodes to be combined, and the nodes to be combined are combined to form a regular distribution curve

The further design of the database data interface management method based on flexible configuration is that the condition for judging whether to continue iteration is: calculating the size of the NMI normalized mutual information value of the first clustering result and the second clustering result. If it is greater than the set threshold, the iteration is stopped to obtain the final attribute combination; wherein, the first clustering result is the previous DBSCAN clustering iteration result, and the second clustering result is the next DBSCAN clustering iteration result of the previous one.

The advantages of the present invention are as follows:

The present invention performs accurate sharding clustering by obtaining the best attribute combination to improve query efficiency. Representative attributes are obtained according to the data change characteristics of the same attribute, and benchmark attributes are obtained according to the overall distribution of the data, and then a bipartite graph is established in different iterative processes. The edge weights of the obtained bipartite graph are calculated, and in the process of obtaining the edge weights according to the benchmark attributes, the similarity between the nodes is accurately obtained, and then the edge weights are accurately obtained, and then the best attribute combination is obtained according to the K-M matching result of the bipartite graph, and the database is sharded and clustered according to the best attribute combination to obtain accurate sharding results, thereby improving the query efficiency of the management system.

Detailed ways

The technical solution of the present invention is described in detail below.

Step 1 of this embodiment: obtaining data of different attributes in the database.

In this case, data with different attributes are obtained from the database, where the management system is mainly used for the management of power data. Therefore, this embodiment is explained with power data, where data with different attributes include user ID, user name, time, power usage category, power consumption, electricity fee, etc., and one piece of data contains data with different attributes. For the convenience of calculation, the data with the same attribute is linearly normalized.

Step 2: Obtain the representativeness and benchmark degree of different attributes, obtain representative attributes and benchmark attributes. In different iterations, construct attribute bipartite graphs to obtain the final attribute combination.

It should be noted that in the traditional data sharding process, by clustering the data, the data locality can be further improved, thereby improving the query efficiency. However, in the process of data sharding clustering, if different attribute data are clustered separately, the data sharding results under different attributes are obtained. However, in these data sharding results, due to the certain connection between the attributes, there will be certain repetitiveness in the different data shards obtained, that is, there are a large number of repetitions between the corresponding certain data shards, which will cause spatial redundancy and reduce the query efficiency. In order to accurately cluster the data, the present invention is expected to achieve accurate clustering by attribute combination, that is, corresponding to clustering from the original single attribute to clustering through multiple attributes, so as to achieve accurate clustering while ensuring query efficiency. In the process of obtaining attribute combination, since the data of different attributes contain different information content of the whole data, and the change of attribute data has different benchmarks for the whole data (that is, corresponding to the change of some attributes in which it is not affected by other attribute data, then this type of attribute data can characterize the benchmark distribution of the overall data), according to these attribute data as the characterization conditions of attribute combination, the accuracy of data clustering can be improved.

It should be further explained that in the process of obtaining the attribute combination method, the bipartite graph processing method is used to perform K-M matching on the bipartite graph, and the corresponding K-M matching result is the appropriate attribute combination result. The attribute combination result in each iteration process is obtained through iteration, and the final attribute combination method is obtained comprehensively.

Step 2-1) Obtain the representativeness and benchmark degree of different attributes, and obtain representative attributes and benchmark attributes.

According to all the data in the database, the data change characteristics of the same attribute reflect the representativeness and benchmark degree of the attribute. For the representativeness of attribute data, the representativeness characterizes the information capacity of this attribute. The calculation method of the representativeness _aj of the jth attribute is:

Where _Nj represents the number of data types of the jth attribute (how many different data values appear); max(N) represents the maximum number of data types of all attributes; _Js ^′ represents the number of all attributes except the jth attribute;

Where _cvs ( _nj ) represents the coefficient of variation of the data value of the sth attribute other than the _jth attribute when the jth attribute data value is the njth data value. This value represents the data distribution of other attributes when the jth attribute value is unique. If the data distribution of other attributes is relatively discrete (large coefficient of variation), it indicates that the representativeness of this value is low.

The representativeness of all attributes is linearly normalized, and the attribute corresponding to the maximum benchmark degree is selected as the benchmark attribute.

For the benchmark degree of attribute data, the benchmark degree represents the overall benchmark change degree of the overall data representing this attribute. The benchmark degree is determined by the overall distribution under this attribute. Therefore, it is necessary to obtain the first truncated data when the representative attributes are the same, and determine the benchmark degree of the attribute in the data. The process of obtaining the first truncated segment data is: arrange all the data in the database, and use the arranged position sequence number as the positioning. According to the above steps, all representative attributes are obtained. Taking a certain representative attribute as an example, the data with the same representative attribute is segmented (it should be noted that after the truncation process, the data is not necessarily arranged continuously), and then the truncated data under this representative attribute is obtained; similar operations can be performed to obtain truncated data under other representative attributes. Among all these truncated data, the intersection of these data is obtained, and these intersections are the first truncated data. In each truncated data, the benchmark degree γ _j of the jth attribute is calculated, and the calculation method is:

In the formula, J _s ^′ represents the number of all attributes other than the jth attribute; M represents the number of all truncated data; R _m (s) represents the Pearson correlation coefficient between the data in the mth truncated data under the sth attribute of other non-jth attributes and the data in the mth truncated data under the jth attribute. In the same truncated data, if the Pearson correlation coefficient between the data of the jth attribute and the truncated data under other non-jth attributes is smaller, it means that the change of the data of the jth attribute will not affect the change of other attribute data, and the corresponding jth attribute has a greater baseline degree.

Similarly, the benchmark levels of all attributes are linearly normalized, and the attribute corresponding to the largest benchmark level is selected as the benchmark attribute.

At this point, the representativeness and benchmark degrees of different attributes are obtained, and representative attributes and benchmark attributes are obtained.

Step 2-2) During different iterations, construct an attribute bipartite graph to obtain the final attribute combination.

According to the above steps, representative attributes and benchmark attributes are obtained. For representative attributes, representative attributes are highly representative. In the corresponding attribute combination, the representative attributes must be included. An attribute bipartite graph is constructed to determine the attribute combination. The bipartite graph is a bipartite graph when the representative attribute is a certain same value (the corresponding bipartite graph has multiple), the left node of the bipartite graph is any non-representative attribute, and the right node is also other non-representative attributes. The same attributes in the connected edges between the left and right nodes are not connected. The corresponding edge weights between the nodes can obtain the attribute bipartite graph. In order to ensure the optimal attribute combination, the iterative method is used for matching, that is, the iterative process is: in the first iteration process, the K-M matching of the bipartite graph is performed. After obtaining the attribute combination of the matching result, the difference between the data after clustering according to the attribute combination and the data after clustering before is calculated. The preset difference threshold is 0.45. If the difference is less than the difference threshold, it indicates that the iteration needs to be stopped. The iterative process is to determine whether the number of attribute combinations needs to be increased, and the difference threshold can be determined according to the specific implementation of the implementer.

The edge weight represents the similarity relationship between the left node and the right node. Since there are many different attributes, when calculating the similarity, we cannot only consider the clustering result based on the attribute corresponding to the left node and the similarity with the clustering result based on the attribute corresponding to the right node, that is, the corresponding clustering result based on a single attribute has different contents on different attributes, so the result obtained by the similarity calculation is inaccurate. Therefore, based on the representative attributes obtained, the present invention takes the benchmark attributes as the overall benchmark, and calculates the data distribution curve of the attributes represented by the left node and the right node when the same representative attributes are used (wherein the data distribution is reflected in the form of a coordinate system).

For example, the dth data value of the representative attribute is selected, and its left node is the vth non-representative attribute node. The horizontal coordinate of the data distribution curve of the left node is the benchmark attribute, and the vertical coordinate is the vth non-representative attribute, which is recorded as the first data distribution curve; the right node is the uth non-representative attribute node, and the horizontal coordinate of the data distribution of the right node is the benchmark attribute, and the vertical coordinate is the uth non-representative attribute, which is recorded as the second data distribution curve. Since the first data distribution curve and the second data distribution curve are discrete, the present invention uses the difference between each data point (data point on the data distribution curve) and the corresponding attribute regular distribution curve, and then calculates the similarity of the difference as the edge weight of the edge weight.

The process of obtaining the regular distribution curve is illustrated by taking the vth non-representative attribute node as an example: the horizontal coordinate of the regular distribution curve is also the benchmark attribute, and the vertical coordinate is the vth non-representative attribute. For the horizontal coordinate, obtain the data values of all benchmark attributes, calculate each existing horizontal coordinate point to determine the corresponding vertical coordinate data value, and the calculation method of the vertical coordinate value _yq corresponding to the qth horizontal coordinate value is:

Where H represents the number of clusters after DBSCAN clustering based on the qth benchmark attribute; W _h represents the number of data types in the hth cluster; P _w represents the frequency value of the wth data type; G _w represents the data value of the wth data type;

In the formula, Indicates the distribution characteristics of the cluster where the wth data type in the hth cluster is located, which is represented by the density between the data in the cluster. The vertical coordinate value is obtained by weighted average, where / > Indicates the weight of the data type. If the cluster where the data is located is relatively scattered, it means that the representation of the data type is relatively random, and the corresponding weight of the data type is relatively small.

According to the obtained ordinate value corresponding to the horizontal coordinate, a relatively continuous distribution curve is obtained, and the distribution curve is fitted to obtain the corresponding regular distribution curve, wherein the regular distribution curve of the left node (the vth non-representative attribute node) is recorded as the first regular distribution curve, and the regular distribution curve of the right node (the uth non-representative attribute node) is recorded as the second regular distribution curve.

Then the difference between the first data distribution curve and the first regular distribution curve, and the difference between the second data distribution curve and the second regular distribution curve are calculated accordingly. When calculating the difference, the ordinate value of the horizontal coordinate of the data point on the first data distribution curve is subtracted from the ordinate value of the horizontal coordinate of the first regular distribution curve. The absolute value of the difference is the difference value of this horizontal coordinate of the left node, which is recorded as the first difference value (including multiple first difference values, one horizontal coordinate corresponds to one). Similar operations are performed to obtain the difference value of the same horizontal coordinate of the right node, which is recorded as the second difference value (also including multiple second difference values, one horizontal coordinate corresponds to one). The ratio of the first difference value and the second difference value of the same horizontal coordinate is obtained by subtracting 1, and the absolute value of the result is the difference value of this horizontal coordinate. The corresponding average of the difference values of multiple horizontal coordinates is recorded as the difference between the left node and the right node. The difference is normalized by an inverse proportional function, and the result is recorded as the edge weight of the two nodes.

Similar to the above operation, the edge weights between other nodes can be obtained. Perform K-M matching on this bipartite graph, and obtain the two nodes corresponding to the largest edge weights in the matching results, which are recorded as the nodes to be combined. Combine the nodes to be combined (similar to a tuple). The condition for judging whether to continue iteration is: the iteration condition is not calculated for the first time. For the second time, DBSCAN clustering is performed on the data according to the attribute combination in this iteration process, which is recorded as the second clustering result. The previous iteration process is the first clustering result. Calculate the difference between the first clustering result and the second clustering result, wherein the NMI (Normalized Mutual Information) of the two clustering results is calculated, the size of the normalized mutual information value is set, and the threshold is set to 0.65. If it is greater than the threshold, stop the iteration and obtain the final attribute combination.

Step 3 performs data sharding and clustering based on the final attribute combination to achieve accurate storage.

According to the above steps, the final attribute combination is obtained, and the data value corresponding to the attribute combination is used as the condition for sharding clustering of the data to obtain the sharding clustering result. The corresponding data of the sharding clustering result is stored, and an index is constructed for easy query.

The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (9)

1. A database data interface management method based on flexible configuration, characterized by comprising: Step 1) Obtain data of different attributes in the database; Step 2) Obtain the representativeness and benchmark degree of different attributes, obtain representative attributes and benchmark attributes; construct attribute bipartite graphs in different iterations to obtain the final attribute combination; Step 3) Data is partitioned and clustered according to the final attribute combination to achieve accurate storage. 2. According to the database data interface management method based on flexible configuration according to claim 1, it is characterized in that in the process of obtaining the attribute combination method in step 2), K-M matching is performed on the bipartite graph, and the corresponding K-M matching result is used as the attribute combination result and the attribute combination result in each iteration process is obtained through iteration to comprehensively obtain the final attribute combination method. 3. The database data interface management method based on flexible configuration according to claim 1, characterized in that in step 2), the representative degree α _j of the j-th attribute is obtained according to formula (1), In formula (1), _Nj represents the number of data types of the jth attribute, max(N) represents the maximum value of the number of data types of all attributes, _Js ′ represents the number of all attributes except the jth attribute; _cvs ( _nj ) represents the coefficient of variation of the data values of the sth attribute other than the _jth attribute when the jth attribute data value is the njth data value. 4. According to the database data interface management method based on flexible configuration as described in claim 3, it is characterized in that in the step 2), the representativeness of all attributes is linearly normalized, and the attribute corresponding to the largest benchmark degree is selected and recorded as the benchmark attribute. 5. The database data interface management method based on flexible configuration according to claim 3 is characterized in that the benchmark degree in step 2) is determined by the overall distribution under the current attribute, specifically: when the representative attributes are the same, the truncated data are obtained, and so on, the truncated data under other representative attributes are obtained. Among all the truncated data, the intersection of the truncated data under all the representative attributes is obtained, and the intersection is the first truncated data. In each truncated data, the benchmark degree γ _j of the jth attribute is calculated according to formula (2), In formula (2), J _′s represents the number of all attributes other than the jth attribute; M represents the number of all truncated data; _Rm (s) represents the Pearson correlation coefficient between the data in the mth truncated data under the sth attribute of other than the jth attribute and the data in the mth truncated data under the jth attribute. 6. According to the database data interface management method based on flexible configuration according to claim 5, it is characterized in that the process of obtaining the first truncated segment data is: all the data in the database are arranged, the arranged position serial number is used as the positioning, the data with the same representative attributes are segmented, and the truncated data under the current tabular attribute is obtained. 7. The database data interface management method based on flexible configuration according to claim 5, characterized in that the process of obtaining the final attribute combination in step 2) is specifically: Based on the representative attributes and benchmark attributes obtained, the attribute combination is determined by constructing an attribute bipartite graph, wherein the bipartite graph is a bipartite graph when the representative attributes are set to the same value, wherein the left node of the bipartite graph is set to any non-representative attribute, and the right node is set to other non-representative attributes, and the same attributes in the edges connected between the left and right nodes are not connected, then the corresponding edge weights between the nodes can be obtained to obtain the attribute bipartite graph; The matching is performed in an iterative manner. The iterative process is as follows: in the first iteration, the K-M matching of the bipartite graph is performed. After the attribute combination of the matching result is obtained, the difference between the data clustered according to the attribute combination and the data clustered before is calculated. If the difference is less than the preset difference threshold, the iteration is stopped. On the basis of the representative attributes obtained, the benchmark attributes are used as the overall benchmark to calculate the data distribution curves of the attributes represented by the left node and the right node when the representative attributes are the same. Select the d-th data value of the representative attribute, whose left node is the v-th non-representative attribute node, the data distribution curve of the left node has the benchmark attribute on the horizontal axis, and the v-th non-representative attribute on the vertical axis, which is recorded as the first data distribution curve; the right node is the u-th non-representative attribute node, the data distribution of the right node has the benchmark attribute on the horizontal axis, and the u-th non-representative attribute on the vertical axis, which is recorded as the second data distribution curve, and the difference between the data point on each data distribution curve and the corresponding attribute regular distribution curve is used, and then the similarity of the difference is calculated as the edge weight of the edge weight. 8. According to the database data interface management method based on flexible configuration as claimed in claim 7, it is characterized in that the acquisition process of the regular distribution curve is specifically as follows: the horizontal coordinate of the regular distribution curve is set as the benchmark attribute, the vertical coordinate is set as the vth non-representative attribute, and each existing horizontal coordinate point is calculated to determine the corresponding vertical coordinate data value, that is, the vertical coordinate value _yq corresponding to the qth horizontal coordinate value is calculated according to formula (3): In formula (3), H represents the number of clusters after DBSCAN clustering based on the qth benchmark attribute; W _h represents the number of data types in the qth cluster; P _w represents the frequency value of the wth data type; G _w represents the data value of the wth data type; Indicates the distribution characteristics of the cluster where the wth data type in the th cluster is located, which is represented by the density between the data in the cluster; the vertical coordinate value is obtained by weighted average, through/> Indicates the weight of the data type; According to the ordinate value corresponding to the obtained horizontal coordinate, a continuous distribution curve is obtained, and the distribution curve is fitted to obtain the corresponding regular distribution curve. The regular distribution curve of the vth non-representative attribute node is recorded as the first regular distribution curve, and the regular distribution curve of the uth non-representative attribute node is recorded as the second regular distribution curve; the ordinate value of the horizontal coordinate of the data point on the first data distribution curve is subtracted from the ordinate value of the horizontal coordinate of the first regular distribution curve, and the absolute value of the difference is obtained to obtain the difference value of this horizontal coordinate of the left node, which is recorded as the first difference value, and the difference value of the same horizontal coordinate of the right node is obtained by analogy, which is recorded as the second difference value, and the absolute value of the result obtained by subtracting the ratio of the first difference value and the second difference value of the same horizontal coordinate from 1 is obtained as the difference value of the current horizontal coordinate, and the corresponding average of the difference values of multiple horizontal coordinates is recorded as the difference between the left node and the right node, and the difference is normalized by an inverse proportional function, and the processing result is recorded as the edge weight of the two nodes, and the above operation is repeated to obtain the edge weights between other nodes; Perform K-M matching on the bipartite graph, obtain the two nodes corresponding to the largest edge weight in the matching result, record them as the nodes to be combined, and combine the nodes to be combined to form a regular distribution curve. 9. According to the database data interface management method based on flexible configuration as described in claim 7, it is characterized in that the condition for judging whether to continue iteration is: calculating the size of the NMI normalized mutual information value of the first clustering result and the second clustering result. If it is greater than the set threshold, the iteration is stopped to obtain the final attribute combination; wherein, the first clustering result is the previous DBSCAN clustering iteration result, and the second clustering result is the next DBSCAN clustering iteration result of the previous one.