Design And Implementation Of Real-time Decision Engine Supporting Stateful Rules

Because real-time decision engines can provide integrated real-time decision-making capabilities from streaming data access,online rule matching and decision policy execution,more and more enterprises are applying them to their business in recent years.More efficient processing time-dependent stateful rules and improving resource scheduling efficiency of decision tasks are the key issues that real-time decision engines need to face in their applications.In this context,a real-time decision engine supporting stateful rules is designed and implemented.This engine achieves real-time decision-making for streaming data by connecting to real-time data streams,matching the content of data streams and executing relevant decision-making strategies according to the matching results.To meet the requirement of rule online update,a dynamic loading mechanism of rules is designed and implemented to support rule online configurable and modifiable.To solve the problem that rule matching algorithm does not support stateful rules,a stateful rule processing method based on S-Rete algorithm is presented.Support for state rules is achieved by adding State Node to the rule network and optimizing the rule compilation and runtime processes.To solve the inefficiency of traditional Flink task resource scheduling based on manual,a Flink task scheduling method based on time series prediction is proposed,which uses improved Long Short-Term Memory artificial neural network which adapted to Flink task scheduling predicts future data traffic based on real historical traffic.Flink tasks dynamically schedules according to the predicted values,which improves task execution efficiency and reduces system resource overhead.This thesis first introduces the research background and current status of real-time decision engine;investigates the related technologies and products of real-time decision engine,and analyses the requirements of real-time decision engine supporting stateful rules;then puts forward a stateful rule processing method based on S-Rete algorithm for stateful rules;and puts forward a Flink task dispatcher based on time series prediction for Flink task scheduling.Then explain the design and implementation of each module and the system architecture.Finally,the validity of this engine is verified through a series of tests.