Causal Inference For Multi-Variable Coupling Time-Series Information

In recent years,causal inference has become one of the hot research fields after machine learning,which provides the interpret ability for future AI and swarm intelligence.In this thesis,we mainly study the inference of causal information flow of multivariate time series,and use the strong coupling time series data to construct causal network to explore the underlying interaction structure between individuals.Therefore,this thesis introduces the optimal causal entropy algorithm,which calculates the conditional mutual information with time delay to inference the causal network,and applies it to the field of collective behavior and transportation.This dissertation is divided into five chapters and the main contents are summarized as follows.In the first chapter,we introduce the background knowledge of causal information flow infer-ence,summarizes the current research situation in this field at home and abroad,and introduces two traditional causal inference algorithms.Then the motivation and derivation of the main work is presented.In the second chapter,we introduce the causal entropy through the basic knowledge of informa-tion theory,and lead to the optimal causality entropy algorithm used in this thesis.The algorithm decomposes the whole causal network structure inference problem into the low-dimensional causal network structure inference problem corresponding to each node in the network.By calculating and comparing the value of causal entropy,we can find the causal parents of a target node,then infer the causal information flow in the network.In the third chapter,we apply the optimal causal entropy algorithm to collective data and explore the underlying interaction mechanism of pigeon group cooperative flight.The revealed causal relationship follows a local interaction mode,and if the individual closer to the mass center and the average velocity direction,it is more influential to others.In the fourth chapter,we apply the optimal causal entropy algorithm to traffic simulation data to mining the implicit causal information flow in the traffic network.Then evaluate the causal impact of intersections and analyse the characteristics of causality between intersections.We find the conclusion that there is less possibility of strong causality between intersections that are further apart.In the last chapter,we make a summary and propose the relevant research work in the future.