A method and system for mining similar information of dam safety multi-point monitoring effect quantity

Abstract

The present invention belongs to the technical field of dam multi-point monitoring, and specifically relates to a method and system for mining similar information of multi-point monitoring effect quantities of dam safety, including: performing time deviation edit distance similarity analysis on the dam multi-point monitoring effect quantities to obtain a time deviation edit distance similarity matrix; converting the time deviation edit distance similarity matrix into a multi-point graph adjacency matrix; jointly representing the multi-point monitoring effect quantity measurement information and the adjacency matrix as a multi-point graph; using a greedy module to maximize the strong similarity set mining method to mine the multi-point graph for strong similarity sets, and obtaining the multi-point graph strong similarity set subgraph division result; according to the strong similarity set subgraph division result, determining all the strong similarity set mining results in the dam multi-point monitoring effect quantity. The present invention realizes the preliminary mining of similar information between the dam multi-point monitoring effect quantities and the dam multi-point monitoring effect quantity strong similarity set mining method.

Description

Technical Field

[0001] The invention belongs to the technical field of dam multi-measurement point monitoring, and in particular relates to a method and system for mining similar information of effect quantities of dam safety multi-measurement point monitoring. Background Art

[0002] In dam projects, in order to understand the changes in the service performance of the dam in detail, a large number of monitoring instruments are often buried to receive the change information of the corresponding monitoring effect quantity at each measuring point and present it in the form of monitoring data. The monitoring information contained in the monitoring effect quantity of multiple measuring points mainly includes two aspects: the measurement information of each measuring point and the similarity information between the monitoring effect quantities. Among them, for the similarity information between the monitoring effect quantities of multiple measuring points of the dam, from different perspectives, it is generally divided into two types of mining methods. One is to measure the similarity between the monitoring effect quantities of two measuring points, and obtain the similarity information under all combinations; the other is to mine the sets with strong similarity relationships in the monitoring effect quantities of each measuring point, and obtain all strong similarity sets. In the analysis process, it is necessary to adopt the corresponding information representation form in order to further conduct the corresponding dam service performance analysis.

[0003] Under the combined effects of environmental and load factors, the service state of a dam will show obvious time-dependent change characteristics, especially for concrete dams, where the values ​​of the monitoring effect quantities at multiple measuring points often have similar change patterns. In the traditional analysis of the change of dam safety monitoring effect quantities, the time series of the monitoring effect quantities at a single measuring point are mainly based on the time characteristics of a single measuring point, without fully considering the similar characteristics of the monitoring effect quantities at different monitoring points of the dam in space, which has limitations such as insufficient representativeness and information conflict. In order to make full use of the similar information between the values ​​of the monitoring effect quantities at the measuring points and deeply mine the implicit information of the dam's state changes, scholars have proposed some methods for analyzing the change laws of the monitoring effect quantities at multiple measuring points. In the application process of these methods, it is necessary to solve the problem of mining similar information between the values ​​of multiple measuring points. In the mining of similar information of the monitoring effect quantities at multiple measuring points, some progress has been made, but the relevant methods still have certain limitations when facing complex monitoring data, and further research on relevant mining methods is needed. At present, a number of valuable research results have been achieved in the analysis and prediction of dam service state. However, due to the large number of measuring points in dam projects and the complexity of the dam operation environment and its working mechanism, the existing methods are insufficient in the analysis and processing capabilities of multi-point monitoring data. Therefore, how to effectively extract the similar information contained in the multi-point monitoring data and apply it to the analysis and prediction of dam service performance is an important research topic in the field of dam engineering. The following issues need to be further explored: (1) The change of dam monitoring effect quantity is extremely complex. There is often a certain time lag between the monitoring effect quantity and the influence quantity. At the same time, many factors such as noise pollution interfere with the dam monitoring effect quantity data. The traditional monitoring effect quantity similarity analysis method is difficult to obtain a good similarity measurement result. Therefore, it is necessary to explore similar information mining and characterization methods between monitoring effect quantities that can adapt to the above situation. (2) During the service of the dam, due to the continuous changes in its own conditions and external environmental conditions, the dam's changing properties will change accordingly, which may cause the similarity relationship between the monitoring effect quantities of each measuring point to change, thereby causing the original similar information mining results to become invalid. In order to ensure the real-time effectiveness of the similar information mining results, it is necessary to study the tracking and analysis methods of the original similar information mining results when the measurement sequence changes. (3) Traditional dam service performance analysis methods are mostly based on the perspective of a single measuring point. Although the monitoring effect information of a single measuring point can reflect the changes in the dam service performance within its local range, it cannot fully reflect the changes in the overall service performance of the project. It also does not consider the similarity between the monitoring effect quantities of each measuring point, which can easily lead to the loss of some key information. Therefore, it is necessary to explore the dam service performance analysis method based on the similar information between the monitoring effect quantities. (4) The dam monitoring effect quantity is affected by many factors, and there are characteristics such as nonlinearity and hysteresis effects. The traditional dam single measuring point monitoring effect quantity prediction method has the disadvantages of insufficient learning ability and the influence of human subjective factors.In response to the above problems, it is necessary to further study the method of constructing a single-point monitoring effect quantity prediction model for dams, and explore the multi-point monitoring effect quantity prediction technology based on similar information between monitoring effect quantities. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to address the deficiencies in the prior art and provide a method and system for mining similar information between multi-point monitoring effect quantities of dam safety. By studying the similarity analysis method of time deviation edit distance, the preliminary mining of similar information between multi-point monitoring effect quantities of dams is achieved, and a graph data representation of multi-point monitoring information is performed. On this basis, a method for mining strong similarity sets of multi-point monitoring effect quantities of dams is constructed.

[0005] Technical solution: The method for mining similar information of dam safety multi-point monitoring effect quantity of the present invention comprises the following steps:

[0006] S1. Perform time deviation edit distance similarity analysis on the monitoring effect of multiple measuring points of the dam to obtain a time deviation edit distance similarity matrix;

[0007] S2, converting the time deviation edit distance similarity matrix into a multi-point graph adjacency matrix;

[0008] S3, representing the multi-measurement point monitoring effect measurement information and the adjacency matrix together as a multi-measurement point graph;

[0009] S4, using the greedy module to maximize the strong similarity set mining method to mine the strong similarity set of the multi-measurement point graph, and obtain the subgraph division result of the strong similarity set of the multi-measurement point graph;

[0010] S5: According to the strong similarity set subgraph division results, all the strong similarity set mining results in the dam multi-measurement point monitoring effect quantity are determined.

[0011] To further improve the above technical solution, S1 includes:

[0012] S101, selecting multi-measurement point monitoring effect size sequences of the same length within the analysis period, and arranging them into a multi-measurement point monitoring effect size sequence matrix;

[0013] S102, normalizing each monitoring effect size sequence in the matrix;

[0014] S103, traversing all the combinations of two measuring points in the matrix, and calculating the TWED distance between each monitoring effect size sequence;

[0015] S104, arranging the TWED distance values ​​into a symmetrical TWED distance similarity matrix.

[0016] Furthermore, the temporal deviation edit distance similarity analysis has three basic edit operations: match, and ,For each editing operation, the corresponding editing distance is defined, including the measured spatial distance and time distance, as well as the penalty constant for the delete operation;

[0017] Before starting the calculation, set the initialization conditions, including: when the sequence length is 0, the distance is 0, and when the delete operation is performed, the distance is infinite;

[0018] Dynamic programming is used to calculate the time-deviation edit distance between two sequences by constructing a cumulative edit distance calculation matrix to record the minimum edit distance from the beginning of the sequence to any position;

[0019] By comparing the edit distances obtained from all edit operations, the minimum edit distance for transforming sequence A into sequence B is found to obtain the time-deviation edit distance.

[0020] Furthermore, two monitoring effect size sequences A and B are defined in the 2D space, with sequence lengths of N and M respectively. represents the pth measured value of the monitoring effect size sequence A and its marked value on the time axis, that is, , represents the qth measured value of the monitoring effect size sequence B and its mark value on the time axis, ;

[0021] The match editing operation refers to: middle Drag and drop segments into the Monitor Effect Sizes sequence middle The matching operation is completed on the segment;

[0022] Said Editing operation refers to: monitoring effect size sequence middle segment, the measurement point Drag and drop to the previous measurement point Complete the deletion operation;

[0023] Said Editing operation refers to monitoring the effect size sequence middle segment, the measurement point Drag and drop to the previous measurement point The deletion operation is complete.

[0024] Further, the S2 includes:

[0025] S201, normalizing each element in the TWED distance similarity matrix, and determining the maximum distance value between two normalized measurement sequences of length L;

[0026] S202, calculating the difference of each element in the normalized TWED distance similarity matrix relative to 1;

[0027] S204, calculating the weight of each element in the adjacency matrix, thereby forming an adjacency matrix of the multi-measurement point graph.

[0028] Further, the S3 includes:

[0029] S301, define a node set V(G) of a multi-measurement point graph, where each measurement point corresponds to a node in the graph, and the node attribute includes the monitoring effect measurement value information of the measurement point;

[0030] S302, using the adjacency matrix obtained in S2 to define the edge set E(G), for each non-zero element in the adjacency matrix , add an undirected edge between nodes i and j, and set the weight of the edge to The value of

[0031] S303, construct a multi-point graph using the node set V(G) and the edge set E(G) , define the multi-point graph as an undirected weighted graph , where V is the node set, representing each measurement point, and E is the edge set, representing the similarity relationship between the measurement points.

[0032] Further, the S4 includes:

[0033] S401, initialize a multi-point graph, take each node as a strong similarity set subgraph, and calculate the initial modularity;

[0034] S402, merging edge-connected strongly similar set subgraph pairs in sequence in the direction from the largest to the smallest modularity increment, and calculating the modularity after merging;

[0035] S403, repeat step S402, and record the subgraph structure and modularity of the strongly similar set after each merging;

[0036] S404, when the entire multi-measurement point graph is merged into a strong similarity set subgraph, stop, and select the strong similarity set subgraph partition result when the modularity is the largest in the process as the optimal strong similarity set mining result.

[0037] The system for implementing the above-mentioned dam safety multi-point monitoring effect quantity similarity information mining method comprises:

[0038] A data preprocessing module is used to receive the original data of multi-measurement point monitoring effect quantity, select the multi-measurement point monitoring effect quantity sequence of the same length within the analysis period, and perform normalization processing on the selected monitoring effect quantity sequence;

[0039] A similarity analysis module is used to calculate the time deviation edit distance between each monitoring effect size sequence using a time deviation edit distance algorithm, and to organize the calculated time deviation edit distance values ​​into a symmetrical time deviation edit distance similarity matrix;

[0040] The adjacency matrix construction module is used to determine the possible maximum distance value between two normalized measurement sequences, normalize each element in the time deviation edit distance similarity matrix, calculate the difference of each element in the normalized time deviation edit distance similarity matrix relative to 1, and calculate the weight of each element in the adjacency matrix according to the set parameters;

[0041] The multi-point graph construction module is used to create a node set of the graph according to the multi-point monitoring effect measurement value information, create an edge set of the graph according to the adjacency matrix, and combine the node set and the edge set into a complete multi-point graph structure;

[0042] The strong similarity set mining module is used to initialize the multi-point graph, take each node as a strong similarity set subgraph, calculate the initial modularity, iteratively merge the strong similarity set subgraph pairs connected by edges, record the strong similarity set subgraph structure and modularity after each merger, and select the strong similarity set subgraph partition result when the modularity is the largest in the process as the optimal result;

[0043] The result output module is used to organize the strong similarity set mining results and generate reports or visual charts containing strong similarity set information.

[0044] Beneficial effects: Compared with the prior art, the advantages of the present invention are as follows: (1) In order to overcome the limitations of traditional similar information extraction methods when facing complex monitoring data, the present invention studies the basic principles of the time deviation edit distance method, and proposes a similarity analysis method for the change of the monitoring effect quantity of multiple measuring points of dam safety. By constructing a time deviation edit distance similarity matrix, the preliminary mining of similar information of the monitoring effect quantity of multiple measuring points of dam is realized. (2) The relevant concept of using graphs to represent the monitoring effect quantity information of multiple measuring points of dam is proposed. On this basis, by studying the method of converting the time deviation edit distance similarity matrix into the adjacency matrix of the multi-measuring point graph, the representation form of the dam safety multi-measuring point monitoring effect quantity graph data based on the adjacency matrix is ​​realized. (3) In order to deeply mine the similarity information between the measured values ​​of multiple measuring points, the concept of strong similarity sets in multiple measuring points of the dam monitoring system is proposed, and based on the relevant theory of community discovery in graph networks, the method of maximizing the strong similarity set mining method of the multi-measuring point graph greedy module is studied, thereby realizing the strong similarity set mining method of the monitoring effect quantity of multiple measuring points of dam safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the similarity information mining method of dam safety multi-measurement point monitoring effect quantity of the present invention;

[0046] Figure 2 It is a flow chart of the method for preliminary mining of similar information between effect measurement values ​​monitored at multiple measurement points. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the embodiments.

[0048] Example 1: Figure 1 As shown, the method for mining similar information of dam safety multi-measurement point monitoring effect quantities provided in this embodiment includes the following steps:

[0049] S1: For the monitoring effect quantity of multiple measuring points of the dam, the time deviation edit distance similarity analysis method is used to carry out preliminary similarity information mining and obtain the TWED distance similarity matrix.

[0050] S2: After weight transformation, the TWED distance similarity matrix is ​​converted into a multi-point graph adjacency matrix.

[0051] S3: The multi-point monitoring effect measurement information and the adjacency matrix are jointly represented as a multi-point graph.

[0052] S4: Through the multi-measurement point graph greedy module to maximize the strong similarity set mining method, the multi-measurement point graph strong similarity set subgraph partitioning result is obtained.

[0053] S5: Arrange all the strong similarity set subgraph partitioning results to obtain all the strong similarity set mining results in the group of multi-measurement point monitoring effect quantities.

[0054] Specifically, in step S1, the time deviation edit distance (TWED) similarity analysis is performed on the monitoring effect quantities of multiple measuring points of the dam. On this basis, a TWED distance similarity matrix is ​​constructed to achieve preliminary mining of similar information between the monitoring effect quantities of multiple measuring points of the dam, providing a basis for subsequent further mining.

[0055] The time-biased edit distance similarity analysis method is based on the editing operation. For two monitoring effect size sequences A and B in a 2D space, the sequence lengths are N and M respectively. represents the pth measured value of the monitoring effect size sequence A and its marked value on the time axis, that is, ; Similarly, represents the qth measured value of the monitoring effect size sequence B and its marked value on the time axis, that is, .

[0056] In order to transform the monitoring effect size sequence A into the monitoring effect size sequence B with a smaller edit distance, three editing operations different from the classical editing operations are adopted, namely, match, and .

[0057] Match editing operation: Monitor effect size sequence middle Drag and drop segments into the Monitor Effect Sizes sequence middle The matching operation is completed on the segment;

[0058] Editing operations: For monitoring effect size series middle segment, the measurement point Drag and drop to the previous measurement point Complete the deletion operation;

[0059] Edit operation: The specific deletion operation is the same as The editing operation differs only in the monitoring effect size sequence middle Completed on paragraph.

[0060] For the above editing operations, the edit distance is measured mainly from the perspectives of measurement space and time. In addition, an additional penalty term is added for the delete operation. The edit distance generated by each editing operation can be expressed as follows:

[0061]

[0062] Where: is the edit distance generated by the corresponding edit operation; For the measured value and The space distance between of norm; is a flexible measure of time distance; A penalty constant for delete operations.

[0063] The goal of the TWED editing operation is to convert the monitored effect size sequence Effect size series Completely superimposed together, the editing process starts from the front end of the two monitoring sequences and proceeds step by step from front to back on the time axis. In this editing process, the cumulative editing distance of each stage is continuously obtained to form a cumulative editing distance calculation matrix, and finally the TWED distance calculation result between the two monitoring effect size sequences is obtained.

[0064] To better describe this process, the monitoring effect size series and The measured values ​​of ,in, Indicates that the discrete time coordinate is between 1 and A monitoring effect size sequence between represents an empty monitoring effect size sequence (zero length); Indicates that the editing process is in progress. and The cumulative edit distance calculated for these two subsequences is and For these two subsequences, there are three situations in the editing process:

[0065] (1) and Based on the editing process of these two subsequences, the monitoring effect size sequence middle Effect size sequence middle Match editing operations between segments;

[0066] (2) and Based on the editing process of these two subsequences, the monitoring effect size sequence middle Segment Editing operations;

[0067] (3) and Based on the editing process of these two subsequences, plus monitoring effect size sequence B Segment Edit operation.

[0068] Different editing processes will produce different edit distance calculation results, so it is necessary to compare the edit distances obtained in the above three cases and take the minimum value as the current and The cumulative edit distance between two subsequences can be expressed by the following formula:

[0069]

[0070] In order to ensure the smooth progress of the calculation process, the initialization conditions need to be set as:

[0071] ,

[0072] ,

[0073] ,

[0074] To facilitate the understanding of the above-mentioned edit distance calculation principle, the effect size sequence is monitored , For example, when the elasticity coefficient v is 0.001, the editing operation penalty constant λ is 1, and the measured value distance norm LP is 2, the TWED distance calculation result is 7.008.

[0075] Based on the similarity analysis results of time deviation edit distance, we propose Figure 2 The basic process of the method for preliminary mining of similar information between multi-point monitoring effect measurement values ​​is shown in the figure. The specific implementation steps are as follows:

[0076] S101, selecting a multi-measurement point monitoring effect size sequence of the same length within the analysis period, and arranging it into a matrix form;

[0077] S102, normalizing each column of monitoring effect size sequence in the monitoring effect size matrix;

[0078] S103, traversing all pairwise combinations in the matrix, and calculating the TWED distances between each monitoring effect size sequence;

[0079] S104, arranging the TWED distance values ​​between each monitoring effect size sequence into a symmetrical TWED distance similarity matrix;

[0080] S105. Based on the TWED distance similarity matrix, preliminary mining results of similar information between the effect measurement values ​​monitored at multiple measurement points are obtained.

[0081] The information of the monitoring effect of multiple measuring points on a dam contains both point features and the relationship between points. This data structure is actually a graph data structure. Therefore, a graph can be used as a joint representation of the two aspects of the monitoring effect of multiple measuring points.

[0082] The multi-point monitoring effect information can be represented as a node set V(G) and an edge set E(G), which can be represented in the form of a graph. The node set V(G) can directly use the measurement information of the multi-point monitoring effect, while the edge set E(G) requires a certain transformation of the similarity relationship between the multi-point monitoring effect measurements. The adjacency matrix is ​​a common representation of the edge set E(G). Therefore, a multi-point monitoring effect information graph representation based on the adjacency matrix is ​​constructed.

[0083] (1) Construction of adjacency matrix of multi-point graph

[0084] For the effect quantity information of multi-point monitoring of dam safety, the established multi-point graph is an undirected weighted graph. is a multi-point undirected weighted graph with n nodes, then the element corresponding to the i-th row and j-th column in the adjacency matrix A of the graph G is It can be expressed as:

[0085]

[0086] In the formula, Represents the weight of the edge.

[0087] For the multi-point graph established by the multi-point monitoring effect information of dam safety, its adjacency matrix has the following properties:

[0088] 1) The multi-point graph is a simple graph, does not include self-loops, and all elements on the main diagonal of its adjacency matrix are 0, that is, .

[0089] 2) The multi-point graph is an undirected graph, and its adjacency matrix is ​​symmetric, that is, ; In addition, the number of non-zero elements in the i-th row (or i-th column) of its adjacency matrix is ​​exactly the degree of the i-th node.

[0090] (2) Determination of each element in the adjacency matrix

[0091] The stronger the similarity between the values ​​of the multi-point measurements of the dam, the greater the weight of the edge in the multi-point graph. However, for the TWED distance between the monitoring effect size sequences obtained by the time-deviant edit distance algorithm, the stronger the similarity between the measurement point values, the smaller the TWED distance value, and the minimum is 0; the worse the similarity, the larger the TWED distance value. Therefore, the TWED distance similarity matrix between the monitoring effect size measurements cannot be directly used as the adjacency matrix of the multi-measurement point graph and needs to be converted. During the conversion process, in addition to ensuring that the stronger the similarity, the larger the edge weight, the similarity relationship weight needs to be normalized to a value in the range of (0~1) to facilitate subsequent applications; in addition, in order to reflect the similarity differences between the measurement points of multiple measurement points, the stronger similarity relationship should be highlighted. The specific steps for converting the TWED distance similarity matrix into the adjacency matrix of the multi-measurement point graph are as follows:

[0092] S201: Determine the possible maximum distance value between two normalized measurement sequences of length L according to the time deviation edit distance algorithm (2L-1), used to compare the elements in the TWED distance similarity matrix Normalization is performed, where is the TWED distance between the monitoring effect size sequences of the i-th measuring point and the j-th measuring point.

[0093] S202, calculating the difference of each element in the normalized TWED distance similarity matrix relative to 1.

[0094] S203, calculate each element in the adjacency matrix , whose value is ,in The value of is determined by the distribution effect of the weights, thereby forming the adjacency matrix of the multi-point graph.

[0095] (3) Create a multi-point map

[0096] By using the constructed adjacency matrix and combining the measured value information of the multi-measurement point monitoring effect quantity, a multi-measurement point diagram can be established to characterize the information changes of the multi-measurement point monitoring effect quantity of dam safety.

[0097] Since dams have the characteristics of continuous changes, in the dam safety monitoring system, there is often a phenomenon that some multi-point monitoring effect measurement values ​​have a strong similarity relationship. These multi-point sets with strong similarity are called strong similarity sets. This strong similarity of strong similarity sets can provide a basis for preprocessing of multi-point monitoring data, analysis and prediction of dam service performance, etc.

[0098] Modularity It can reflect the difference between the tightness of the connection in the strong similarity set subgraph and the random situation. It is a metric method to evaluate the quality of the strong similarity set mining result of a multi-point graph. The larger the modularity, the tighter the strong similarity set subgraph partition result is, and the more reasonable the strong similarity set mining result is. In the actual partitioning process, The highest value usually occurs at The calculation expression is as follows:

[0099]

[0100] In the formula, For Node and nodes The weight of the edges between them. When the network is an unweighted graph, the weight of all edges can be regarded as 1; For all nodes The sum of the weights of the connected edges (nodes in the case of unweighted graphs) degree); For Node The subgraph of the strongly similar set to which it belongs; is the sum of the weights of all edges (the number of all edges in the graph for an unweighted graph); where For Node The probability of connecting to any node, because The sum of the weights of the connected edges is , so in the random case the node and The weight on the edge is Furthermore, the calculation formula of modularity can be simplified as follows:

[0101]

[0102] In the formula, is a strongly similar set of subgraphs The sum of the weights of the edges within; is a subgraph with strong similarity set in random case The sum of the weights of the edges connecting the nodes in .

[0103] This formula can be further simplified as:

[0104]

[0105] In the formula, is a strongly similar set of subgraphs The weighted average of the edges within ; is a subgraph with strong similarity set in random case The average weight of the edges connecting the nodes in the .

[0106] A method for mining strong similarity sets by greedy modular maximization in multi-measurement point graphs. This method is based on the greedy idea of ​​continuously solving local optimal solutions. It divides each node in the multi-measurement point graph. First, each node is formed into a separate strong similarity set subgraph. On this basis, the strong similarity set subgraph that maximizes the value of modularity Q is merged until all nodes in the graph belong to the same strong similarity set. The above analysis process is carried out from bottom to top, and the whole process obtains a tree graph, that is, the leaf nodes of the tree represent the nodes in the multi-measurement point graph, and each layer of the tree corresponds to a specific division of the network. Finally, based on the selection principle of the maximum modularity value, the corresponding subgraph division is found from all hierarchical divisions of the tree graph as the effective division result of the strong similarity set subgraph. The basic steps to implement this method are:

[0107] S401, initialize a multi-point graph, take each node as a strong similarity set subgraph, and calculate the modularity Q0 at this time.

[0108] S402 , merge the strongly similar subgraph pairs connected by edges in the direction of the largest to the smallest modularity increment ΔQ, and calculate the modularity Qi after the merger.

[0109] S403, repeat 402, and record the strong similarity set subgraph structure and modularity Qi after each merging.

[0110] S404, until the entire multi-measurement point graph is merged into a strong similarity set subgraph, the strong similarity set subgraph partitioning result when the modularity Qi is the largest in the process is used as the optimal strong similarity set mining result.

[0111] Through the greedy module maximization strong similarity set mining method, the multi-measurement point graph is divided into multiple strong similarity set subgraphs with close internal connections. The dam measurement points corresponding to each node in the same strong similarity set subgraph form a group of strong similarity sets. By sorting out all the strong similarity set subgraph division results, all the strong similarity set mining results in the monitoring effect quantity of this group of multi-measurement points can be obtained.

[0112] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for mining similar information of dam safety multi-point monitoring effect quantity, characterized in that: The steps include: S1. Perform time deviation edit distance similarity analysis on the monitoring effect of multiple measuring points of the dam to obtain the time deviation edit distance similarity matrix; The S1 includes: S101, selecting multi-measurement point monitoring effect size sequences of the same length within the analysis period, and arranging them into a multi-measurement point monitoring effect size sequence matrix; S102, normalizing each monitoring effect size sequence in the matrix; S103, traversing all the combinations of two measuring points in the matrix, and calculating the TWED distance between each monitoring effect size sequence; S104, arranging the TWED distance values ​​into a symmetrical TWED distance similarity matrix; The temporal deviation edit distance similarity analysis has three basic edit operations: match, and ,For each editing operation, the corresponding editing distance is defined, including the measured spatial distance and time distance, as well as the penalty constant for the delete operation; Before starting the calculation, set the initialization conditions, including: when the sequence length is 0, the distance is 0, and when the delete operation is performed, the distance is infinite; Dynamic programming is used to calculate the time-deviation edit distance between two sequences by constructing a cumulative edit distance calculation matrix to record the minimum edit distance from the beginning of the sequence to any position; By comparing the edit distances obtained from all edit operations, the minimum edit distance that transforms sequence A into sequence B is found to obtain the time-deviation edit distance; Define two monitoring effect size sequences A and B in 2D space, with sequence lengths N and M respectively. represents the pth measured value of the monitoring effect size sequence A and its marked value on the time axis, that is, , represents the qth measured value of the monitoring effect size sequence B and its mark value on the time axis, ; The match editing operation refers to: Drag and drop segments into the Monitor Effect Sizes sequence middle The matching operation is completed on the segment; Said Editing operation refers to: monitoring effect size sequence middle segment, the measurement point Drag and drop to the previous measurement point Complete the deletion operation; Said Editing operation refers to monitoring the effect size sequence middle segment, the measurement point Drag and drop to the previous measurement point Complete the deletion operation; S2, converting the time deviation edit distance similarity matrix into a multi-point graph adjacency matrix; The S2 includes: S201, normalizing each element in the TWED distance similarity matrix to determine the maximum distance value between two normalized measurement sequences of length L; S202, calculating the difference of each element in the normalized TWED distance similarity matrix relative to 1; S204, calculating the weight of each element in the adjacency matrix, thereby forming an adjacency matrix of the multi-measurement point graph; S3, representing the multi-measurement point monitoring effect measurement information and the adjacency matrix together as a multi-measurement point graph; S4, using the greedy module to maximize the strong similarity set mining method to mine the strong similarity set of the multi-measurement point graph, and obtain the subgraph division result of the strong similarity set of the multi-measurement point graph; S5: According to the strong similarity set subgraph division results, all the strong similarity set mining results in the dam multi-measurement point monitoring effect quantity are determined.

2. The method for mining similar information of dam safety multi-point monitoring effect quantity according to claim 1 is characterized in that: The S3 includes: S301, define a node set V(G) of a multi-measurement point graph, where each measurement point corresponds to a node in the graph, and the node attribute includes the monitoring effect measurement value information of the measurement point; S302, using the adjacency matrix obtained in S2 to define the edge set E(G), for each non-zero element in the adjacency matrix , add an undirected edge between nodes i and j, and set the weight of the edge to The value of S303, construct a multi-point graph using the node set V(G) and the edge set E(G) , define the multi-point graph as an undirected weighted graph , where V is the node set, representing each measurement point, and E is the edge set, representing the similarity relationship between the measurement points.

3. The method for mining similar information of dam safety multi-point monitoring effect quantity according to claim 2 is characterized in that: The S4 includes: S401, initialize a multi-point graph, take each node as a strong similarity set subgraph, and calculate the initial modularity; S402, sequentially merging edge-connected strongly similar set subgraph pairs in the direction of the largest to smallest modularity increment, and calculating the modularity after merging; S403, repeat step S402, and record the subgraph structure and modularity of the strongly similar set after each merging; S404, when the entire multi-measurement point graph is merged into a strong similarity set subgraph, stop, and select the strong similarity set subgraph partition result when the modularity is the largest in the process as the optimal strong similarity set mining result.

4. A system for implementing the method for mining similar information of dam safety multi-point monitoring effect quantities as described in claim 1, characterized in that: include: A data preprocessing module is used to receive the original data of multi-measurement point monitoring effect quantity, select the multi-measurement point monitoring effect quantity sequence of the same length within the analysis period, and perform normalization processing on the selected monitoring effect quantity sequence; A similarity analysis module is used to calculate the time deviation edit distance between each monitoring effect size sequence using a time deviation edit distance algorithm, and to organize the calculated time deviation edit distance values ​​into a symmetrical time deviation edit distance similarity matrix; The adjacency matrix construction module is used to determine the possible maximum distance value between two normalized measurement sequences, normalize each element in the time deviation edit distance similarity matrix, calculate the difference of each element in the normalized time deviation edit distance similarity matrix relative to 1, and calculate the weight of each element in the adjacency matrix according to the set parameters; The multi-point graph construction module is used to create a node set of the graph according to the multi-point monitoring effect measurement value information, create an edge set of the graph according to the adjacency matrix, and combine the node set and the edge set into a complete multi-point graph structure; The strong similarity set mining module is used to initialize the multi-point graph, take each node as a strong similarity set subgraph, calculate the initial modularity, iteratively merge the strong similarity set subgraph pairs connected by edges, record the strong similarity set subgraph structure and modularity after each merger, and select the strong similarity set subgraph partition result when the modularity is the largest in the process as the optimal result; The result output module is used to organize the strong similarity set mining results and generate reports or visual charts containing strong similarity set information.

Images