Causal Inference On Additive Model With Confounders

Causal inference is a discipline developed on the basis of the statistics department,which specializes in the causal relationship between things.The development of causal inference theory has gone through the process of deterministic causality,empirical theory of causality,and probability theory of causality.Its research methods have also gone from low-dimensional to high-dimensional,from non-core models to models,from simple data.To the development of complex data.The research content of causal inference is divided into two categories,one is the study of causal effect,and the other is the study of causality.The statistical inference of causal effects mainly discusses the identifiability of causal effects and the statistical inference of causal effects.This paper gives a comprehensive review of the development history of the various models of causal inference and the research status at home and abroad,especially the causal inference under mixed conditions.On this basis,a linear additivity model and a nonlinear additivity model are constructed for the confounding data,and a new method for causal inference under the two models is proposed.Specifically,this thesis carries out four parts of theoretical research work: The first part,based on the expression of different path structures in causality based on confounding factors,proposes a linear additivity regression model and a nonlinear additive regression model,so that it can not only It is suitable for the identification of confounding factors under linear relations,and also for the identification of confounding factors under nonlinear relationships.The second part is to estimate the unknown parameters and unknown functions in the two models.The third part is to carry out the two models.Hybrid factor identification,applying the definition of confounding factors and linear trace method,determining the causal relationship by defining the asymmetry relationship between cause and result,establishing a set of algorithms for identifying confounding factors;the fourth part,using hybrid control of the two models,using The solution of the model to explain the elimination of the confounding effect.After theoretical research,the proposed method is statistically simulated to verify the rationality of the proposed method,and a proof or proof summary is given for the important theorem.Finally,an empirical study was carried out to apply the proposed method to explore the possible causes of solar photovoltaic power generation changes and to identify confounding factors that may have a bias effect on the causal relationship.The results of simulation analysis and empirical research show that the hybrid identification method and the hybrid control method based on the linear additivity regression model and the nonlinear additive regression model have achieved excellent results and can be well applied to the hybrid.Causal analysis under the data.Finally,the research of the full text is summarized and the future work is prospected,and the next research work is pointed out.