Extended Belief Rule Base Parameter and Structure Optimization Based on Similarity Measure

The extended belief rule base (EBRB) combines the characteristics of knowledge structure and evidence reasoning, and can quantitatively describe the uncertainty and incompleteness of data. However, the original model will activate all rules whose activation weight is greater than zero, which greatly affects the inference results of the EBRB model. At the same time, the initial parameter settings also have different effects on the output of the model. For example, the accuracy of classification is affected by interval division, meanwhile the setting of rule weights and attribute weights will affect the output of EBRB. In response to the above problems, this paper proposes an EBRB structure and parameter optimization model based on similarity, and applies this model to the health assessment of lithium-ion batteries. EBRB optimized based on similarity measures uses a data-driven approach to convert rule sets into rules, uses multiple kernel maximum mean discrepancies (MKMMD) to calculate the similarity between the test data set and the corresponding rules, and selects rules with high similarity for inference output. Through comparison with traditional methods, the effectiveness of the proposed method is verified.