Research On Ensemble Learning Method Base On Multiple Fusion Weighted Evidential Reasoning Rules

In recent years,integrated learning method has been widely concerned in all walks of life.Its application covers credit risk assessment,medical image recognition,intelligent recognition system,photovoltaic power system,machinery industry and wind power.For example,in the field of credit risk assessment,ensemble learning can effectively combine multiple base classifiers to improve the accuracy of predicting loan default rate,so as to improve the accuracy and efficiency of loan approval.In the field of medical image recognition,ensemble learning can fully mine the effective information of base classifier and improve the accuracy of tumor detection.In the field of intelligent recognition system,integrated learning method can also be used in face recognition,speech recognition,text classification and other aspects,improve the application effect of intelligent recognition technology;In the field of photovoltaic power system,integrated learning can predict the output power of photovoltaic power generation with higher precision,which lays a foundation for optimizing the operating efficiency of photovoltaic power system.In the field of machinery industry,the integrated learning method can be used to predict the remaining life of mechanical parts and fault warning,so as to improve the maintenance efficiency of mechanical equipment and reduce the maintenance cost.In the field of wind power,integrated learning can accurately predict wind speed and power output of wind farms,providing guidance for improving the power generation efficiency of wind farms and ensuring the economy of power generation.However,the current ensemble learning methods mostly adopt the voting fusion strategy,which is difficult to mine the effective information of subclassifiers and can not effectively reflect the relationship between different classifiers.To solve these problems,researchers have proposed some new integrated learning methods.For example,ensemble learning based on weighted fusion strategy can be weighted according to the performance of base classifiers,so as to better reflect the differences between different classifiers.Based on meta-learning ensemble learning,this method can improve the combinatorial effect of classifiers by learning the relationship between classifiers.However,compared with the above research methods,Evidential Reasoning rule(ER rule)is proposed in this paper as a combination strategy of integrated learning.As a semi-quantitative expert system,ER rule can effectively deal with the uncertainty in multi-source information.However,the prediction results of different classifiers are not the same,which contains a large amount of uncertain information.The use of ER rules can effectively deal with such uncertain problems,fully excavate the internal relations between classifiers,and improve the combination effect of classifiers.In addition,the reasoning process of ER rules is clear and transparent,and its results are interpretable and traceable.However,the final evaluation effect of the integrated learning method based on ER rules depends on the weight distribution of different combination strategies,and the setting of evidence weight will greatly affect the accuracy and stability of the model.Therefore,in practical application,it is necessary to select a reasonable combination strategy and weight setting method according to specific problems.At the same time,it is also necessary to pay attention to the interpretability of the model,so as to improve the comprehensibility of the model as much as possible,so that it can be better applied to actual scenarios.This paper proposes a new ensemble learning method based on multiple fusion weighted evidential reasoning rules,and constructs an ensemble learning framework for data fusion and decision mapping.The framework takes the weight of evidence,confidence and characteristic data of each classifier as input,and the integration result as output.At the same time,the subjective weight and objective weight of base classifiers are set respectively by relying on qualitative knowledge and quantitative information,and a new multi-fusion weighting method is proposed,which combines the subjective weight and objective weight as the final evidence weight.This method is more scientific and reliable than the single weight.Finally,multiple data sets are used to verify the validity of the proposed integration strategy and weight setting method.The experimental results show that the ensemble learning method proposed in this paper based on ER rules and considering the diversity weighting has good performance.It not only improves the model generalization ability,but also significantly improves the instability defect of single weight,and further excavates the internal relationship of data,which has a certain interpretability.In order to apply the proposed method to practice,this paper develops an integrated learning data acquisition system based on evidential reasoning rules.The system is embedded with ER core algorithm and fusion empowerment core algorithm,and finally forms a complete process from theoretical innovation to case verification to application implementation with the above content.To sum up,in view of the problems that inheritance strategy in traditional ensemble learning is difficult to excavate effective information of classifiers,cannot fully reflect the relationship between classifiers,and inaccurate and unstable ensemble learning results caused by the setting of different evidence weights,this paper proposes the integration strategy based on ER rules and the diversity weighting method,and makes a detailed theoretical analysis of the proposed methods.The classification of multiple data sets is verified.Finally,the method is successfully applied from theory to practice through the development of the system.