Multi-source fuzzy information fusion method and system based on attention mechanism and graph correlation analysis

Abstract

The application discloses a multi-source fuzzy information fusion method and system based on an attention mechanism and graph correlation analysis, and belongs to the technical field of artificial intelligence and information fusion. The method comprises the following steps: constructing a multi-source fuzzy decision information system; utilizing the attention mechanism to adaptively weight and fuse a plurality of fuzzy information sources, so as to obtain a fusion information system; calculating the importance of attributes and attribute correlation in the fusion information system, and constructing a weighted attribute correlation graph; calculating the global correlation importance of the attributes based on matrix power series propagation; obtaining an attribute ranking result according to the global correlation importance of the attributes, and obtaining a final correlation fusion result according to the attribute ranking result. The application utilizes the attention mechanism to realize adaptive learning of the weights of different information sources, and comprehensively evaluates the direct correlation and indirect correlation among the attributes by combining the attribute importance, the attribute correlation and the graph correlation propagation mechanism, so that the effective fusion of multi-source fuzzy information is realized, the accuracy and stability of the fusion result are improved, and the application can be widely used in the fields of data analysis, pattern recognition, intelligent decision and information processing.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence, information fusion and uncertain information processing technology, and specifically relates to a multi-source fuzzy information fusion method and system based on attention mechanism, fuzzy neighborhood rough set and graph association analysis. Background Technology

[0002] Multi-source information fusion is a key technology that comprehensively processes data from multiple data sources, sensors, or information systems, fully utilizing the complementary information between these sources to obtain more accurate and reliable decision-making results. In real-world environments, data acquisition is easily affected by factors such as equipment differences, environmental noise, and missing information, resulting in multi-source data often exhibiting ambiguity, uncertainty, and redundancy. Existing multi-source information fusion methods mostly employ fixed fusion rules or uniform weighting strategies, making it difficult to accurately reflect the differences in importance and reliability among different information sources and susceptible to interference from low-quality data sources. Furthermore, the fusion results often contain strong attribute correlations and information redundancy, impacting subsequent analysis and decision-making effectiveness. Summary of the Invention

[0003] The purpose of this invention is to overcome the problems existing in multi-source fuzzy information fusion methods, such as fixed weight allocation of information sources, difficulty in accurately representing the importance and reliability of different information sources, and high redundancy of fusion results. This invention provides a multi-source fuzzy information fusion method based on attention mechanisms and graph association analysis. This method can adaptively learn the importance of different information sources, fully explore the correlations between attributes, and achieve efficient fusion of multi-source fuzzy information, thereby improving the accuracy of subsequent decision analysis and classification.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A multi-source fuzzy information fusion method based on attention mechanism and graph association analysis includes: constructing a multi-source fuzzy decision information system; using attention mechanism to fuse multi-source information; calculating attribute importance based on fuzzy neighborhood rough set; calculating attribute correlation based on Spearman rank correlation coefficient; constructing attribute association graph; using matrix power series for global association propagation; and obtaining the final fusion result.

[0005] This technical solution achieves adaptive allocation of the importance of different information sources through an attention mechanism, avoiding the problem of insufficient information utilization caused by the fixed weights used in traditional fusion methods; at the same time, it utilizes a graph association propagation mechanism to mine the global association relationships between attributes, thereby obtaining more accurate and compact fusion results.

[0006] Furthermore, the adaptive weighted fusion of multiple fuzzy information sources using an attention mechanism includes: Represent multiple fuzzy information sources as multi-channel input features; The weight coefficients corresponding to each information source are learned through an attention network; Normalize the weighting coefficients; The features of each information source are weighted and fused according to the normalized weight coefficients to obtain the initial fusion result.

[0007] Furthermore, the technical solution can dynamically adjust the fusion weights according to the contribution of different information sources to the decision-making results, avoiding fusion bias caused by fixed weight strategies and improving the accuracy and reliability of the fusion results.

[0008] Furthermore, the attention network includes: Stack the attribute matrices corresponding to each information source according to the information source dimension; Perform a linear mapping on the stacked multi-source features; Learning the latent relationships between different information sources through nonlinear activation functions; The attention weights for each information source are generated using the Softmax function. The information sources are weighted and summed according to the attention weights to obtain the fusion result.

[0009] Furthermore, in multi-source fuzzy information systems, the degree of contribution of different information sources to the decision results varies significantly. Traditional average fusion or fixed weight fusion cannot accurately reflect this difference, while attention mechanism can automatically learn the importance of different information sources and realize dynamic representation of the contribution of information sources.

[0010] Furthermore, the calculation of attribute importance based on fuzzy neighborhood rough sets includes: Constructing fuzzy neighborhood relationships between objects based on attribute values; Construct a fuzzy positive region based on fuzzy neighborhood relationships; The degree of dependence of the attribute on the decision attribute is calculated based on the fuzzy positive domain. The degree of dependence is used as the importance of the corresponding attribute.

[0011] This further technical solution can effectively characterize the contribution of attributes to decision-making outcomes.

[0012] Furthermore, the calculation of attribute correlation based on Spearman's rank correlation coefficient includes: Sort the attribute values; Calculate the rank difference between any two attributes; Calculate the corresponding Spearman rank correlation coefficient based on the rank difference. The degree of correlation between attributes is obtained based on the Spearman rank correlation coefficient. Determine the non-redundant relationships between attributes based on their degree of correlation.

[0013] This further technical solution can identify highly relevant attributes and reduce the impact of redundant information on the fusion results.

[0014] Furthermore, the construction of the attribute association graph includes: Use attributes as graph nodes; Calculate the relevance importance of attributes based on their importance. Calculate attribute associations based on non-redundant relationships between attributes; Determine the graph edge weights based on attribute relevance and attribute association. Construct a weighted undirected attribute association graph.

[0015] This further technical solution can simultaneously characterize the importance of attributes and the correlation between attributes.

[0016] Furthermore, the global correlation propagation analysis includes: Construct an adjacency matrix based on the attribute association graph; Perform matrix power series propagation calculation on the adjacency matrix; The global correlation importance of each attribute is obtained based on the propagation result of the matrix power series; The attributes are sorted according to the global association importance. The final fusion result is obtained based on the sorting results.

[0017] This further technical solution not only considers the direct relationships between attributes, but also can explore the indirect relationships between attributes, thereby improving the accuracy of attribute evaluation results.

[0018] Furthermore, an attenuation factor is introduced during the propagation of the matrix power series.

[0019] The attenuation factor is used to control the range of information diffusion in the higher-order propagation process and to ensure the convergence of the propagation process.

[0020] This further technical solution can improve the stability of the propagation process and reduce the impact of noise association on the propagation results.

[0021] Compared with existing technologies, this invention has at least the following beneficial effects: it utilizes an attention mechanism to adaptively learn the importance of different information sources, achieving adaptive fusion of multi-source fuzzy information; it combines fuzzy neighborhood rough sets and Spearman rank correlation coefficients, considering both attribute importance and attribute redundancy, thus improving the discriminative ability of the fusion results; it constructs an attribute association graph and uses matrix power series for global association propagation, comprehensively considering both direct and indirect associations between attributes, thereby improving the accuracy of attribute evaluation results; and it combines attention fusion with graph association analysis to achieve effective fusion of multi-source fuzzy information, obtaining more compact and reliable fusion results. Attached Figure Description

[0022] Figure 1 This is a flowchart of an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the structural framework of an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention.

[0025] Figure 1 This is a flowchart of an adaptive attention-based association fusion method for a multi-source fuzzy information system according to an embodiment of the present invention. The following is in conjunction with... Figure 1 The specific implementation process of this invention will be described. For example... Figure 1 As shown, the adaptive attention-based association fusion method for the multi-source fuzzy information system includes the following steps: S1 constructs a multi-source fuzzy decision information system; Let a multi-source fuzzy decision information system be represented as:

[0026] in For a collection of objects, For the first A set of conditional attributes for an information source. For the set of decision attributes, Information function, Indicates the number of information sources; To uniformly describe attribute information from different information sources, each information source is mapped to a unified feature tensor:

[0027] in, The unified feature tensor represents the attribute dimension corresponding to a single information source and serves as the input data for the multi-source information fusion model.

[0028] S2 utilizes the attention mechanism to construct an attention fusion module, learns the importance weights of each information source, and obtains a fused information system.

[0029] For any object x, its corresponding multi-source feature representation is as follows:

[0030] First, the input features are transformed in dimension:

[0031] Subsequently, latent correlation features between different information sources are extracted using a linear mapping layer and a nonlinear activation function:

[0032] in This is the first layer weight matrix. It is the bias vector; Based on this, the attention score is calculated:

[0033] in This is the first layer weight matrix. It is the bias vector; use The function normalizes the attention score:

[0034] Obtain the attention weights corresponding to each information source.

[0035] The information sources are weighted and fused according to the attention weights to obtain the fused representation:

[0036] in, This represents element-wise multiplication. Representation Object In the Feature representation in an information source; The above process can automatically learn the importance of different information sources, achieve adaptive allocation of information source weights, and improve the accuracy and robustness of the fusion results.

[0037] S3, calculate attribute importance and attribute correlation, and construct a weighted attribute association graph.

[0038] After obtaining the fused information system, fuzzy neighborhood relationships between objects are constructed based on attribute values, and the corresponding positive neighborhood domains are calculated.

[0039] For any attribute Based on its set of decision attributes The importance of the positive domain contribution is defined by the attribute:

[0040] in The neighborhood radius, Representing attributes Regarding decision attributes The positive neighbor region; For any two attributes The correlation between attributes was calculated using the Spearman rank correlation coefficient:

[0041] in, Representation Object In attributes With attributes rank difference on Indicates the number of samples; Define attribute discrimination ability based on attribute importance:

[0042] The degree of non-redundancy is defined based on attribute correlation:

[0043] Further determine the graph edge weights:

[0044] in, This represents the balance parameter, used to coordinate discrimination capability and redundancy suppression capability; Using attributes as graph nodes and the relationships between attributes as graph edges, construct a weighted attribute association graph and establish the corresponding adjacency matrix:

[0045] S4 calculates the global association importance of attributes based on matrix power series propagation.

[0046] The adjacency matrix C obtained in step S3 is used for matrix power series propagation analysis.

[0047] Let the length be The cumulative path weight is:

[0048] The length is The total path impact is expressed as:

[0049] Define the local influence of a node:

[0050] in This represents the set of nodes in the graph.

[0051] The global importance of an attribute is obtained by combining the propagation results of different orders:

[0052] in Represents the attenuation factor, satisfying Representing the adjacency matrix spectral radius; The global importance of each attribute is obtained through matrix power series propagation.

[0053] S5: Obtain the attribute ranking results based on the global association importance of the attributes.

[0054] Based on the global association importance of the attributes obtained in step S4, all attributes are sorted to obtain the attribute importance ranking sequence.

[0055] The sorting sequence takes into account both the attribute's own discriminative ability and the direct and indirect relationships between attributes.

[0056] S6: Obtain the final association and fusion result based on the attribute sorting results.

[0057] Based on the attribute sorting results obtained in step S5, the target attribute set is obtained.

[0058] The target attribute set is output as the final association fusion result for subsequent classification decisions, multi-source information analysis, or other data processing tasks.

Claims

1. A multi-source fuzzy information fusion method based on attention mechanism and graph association analysis, characterized in that, include: S1, Construct a multi-source fuzzy decision information system; S2, utilizes the attention mechanism to construct an attention fusion module, learns the importance weights of each information source, and performs adaptive weighted fusion of multiple information sources to obtain a fused information system; S3, calculate the attribute importance and attribute relevance based on the fused information system, and construct a weighted attribute association graph based on the attribute importance and attribute relevance; S4, perform matrix power series propagation calculation on the weighted attribute association graph to obtain the global association importance of the attributes; S5. Obtain the attribute ranking result based on the global association importance of the attribute; S6. Obtain the final association and fusion result based on the attribute sorting result.

2. The multi-source fuzzy information fusion method according to claim 1, characterized in that, Step S2 includes: Represent multiple fuzzy information sources as multi-channel input features; Perform a linear mapping on the multi-channel input features; Using nonlinear activation functions to learn the potential correlations between different information sources; The attention weights corresponding to each information source are obtained through the Softmax function; The information sources are weighted and fused according to the attention weights to obtain a fused information system.

3. The multi-source fuzzy information fusion method according to claim 1, characterized in that, The calculation of attribute importance in step S3 includes: Constructing fuzzy neighborhood relationships between objects based on attribute values; Construct a positive neighborhood region based on the fuzzy neighborhood relationship; Calculate the degree of dependence of each attribute on the decision attribute based on the positive neighbor region; The importance of the corresponding attribute is obtained based on the degree of dependence.

4. The multi-source fuzzy information fusion method according to claim 1, characterized in that, The calculation of attribute correlation in step S3 includes: Sort the attribute values; Calculate the rank difference between any two attributes; Calculate the Spearman rank correlation coefficient based on the rank difference; The degree of correlation between attributes is obtained based on the Spearman rank correlation coefficient.

5. The multi-source fuzzy information fusion method according to claim 1, characterized in that, The construction of the weighted attribute association graph in step S3 includes: Use attributes as graph nodes; Determine the attribute discrimination ability based on attribute importance; Determine the degree of attribute non-redundancy based on attribute correlation; The weights of graph edges are determined based on the attribute discrimination ability and the degree of attribute non-redundancy. A weighted attribute association graph is constructed based on the graph edge weights.

6. The multi-source fuzzy information fusion method according to claim 1, characterized in that, Step S4 includes: Construct an adjacency matrix based on the weighted attribute association graph; Perform matrix power series propagation calculation on the adjacency matrix; The local correlation importance of each attribute is obtained based on the propagation results of different orders; The global correlation importance of attributes is obtained by combining the propagation results of different orders.

7. The multi-source fuzzy information fusion method according to claim 6, characterized in that, The attenuation factor is introduced during the propagation of the matrix power series to control the information diffusion range in the higher-order propagation process and to ensure the convergence of the propagation process.

8. A multi-source fuzzy information fusion system based on attention mechanism and graph association analysis, characterized in that, include: A multi-source fuzzy information system construction module is used to construct multi-source fuzzy decision information systems; The attention fusion module is used to learn the importance weights of each information source and obtain a fused information system. The attribute association graph construction module is used to calculate attribute importance and attribute correlation and construct a weighted attribute association graph; The global association propagation module is used to perform matrix power series propagation calculations and obtain the global association importance of attributes; The attribute sorting module is used to obtain attribute sorting results based on the global association importance of attributes. The association fusion result output module is used to obtain the final association fusion result based on the attribute sorting results.

Images