Research On Data Pre-deployment Based On Rule Inference Engine In Smart Environment

In the smart environment system based on wireless sensor network,the automatic control of equipment depends on the inference process of matching the sensed environment information with the user-defined rules.With the development of wireless sensor networks and the increasing complexity of business logic,the scale of sensor data and rule sets in smart environments has greatly increased.However,the rule matching efficiency of the traditional production inference system is inefficient,and due to the large amount of sensor data,the amount of real-time feature calculation during rule matching is pretty large,which reduces the real-time performance of inference.At the same time,the limited memory resources of edge devices cannot cope with such a huge amount of sensor data.In addition,the needs of new users and the new needs of users pose challenges to the rigid business logic of traditional production inference systems.In view of the problems above,this thesis designs a node intelligence hierarchical model and a rule-based inference engine-based Data Pre-Deployment Scheme(DPDS)suitable for wireless sensor networks,which specifically includes the following content:(1)Designed and proposed a node intelligent hierarchical model,which replaces the static program code encapsulation in the original working mode by using a dynamic script configuration process to achieve the purpose of dynamic allocation of hardware and software resources in the system,and decoupling of business requirements and hardware devices,and then improve the flexibility of the system and realize the rapid deployment of the system for different businesses.(2)Designed and proposed the DPDS architecture,which is mainly composed of the rule parsing and preprocessing module,rule network and Light-weight Characteristic Table(LCT).Using the rule parsing and preprocessing modules to parse the rule set to obtain atomic conditions and statistical units,rule network and LCT is constructed.LCT precompute the features and pre-stores the feature values,so that the rule network could be able to directly references the feature values in the LCT during inference without real-time feature calculations,which significantly improves efficiency and real-time of the inference process.In addition,taking advantage of the static predictability of system memory footprint,the LCT Pre-Deployment Plan was designed,which enables the memory footprint during the inference period to adapt to the system's memory resources.Finally,taking the time delay estimation of sound source localization based on Time Difference of Arrival(TDOA)as an example,DPDS is applied to the time delay estimation.(3)By comparing with the traditional production reasoning system and the reasoning system using the RETE algorithm,it theoretically proves the superiority of DPDS in terms of time complexity and space complexity.A comparative experiment was designed to compare the pros and cons of the inference system between DPDS and traditional production,using the RETE algorithm and the RETE algorithm using fuzzy processing and B+ tree structure under different number of rules,different data flow rates and different maximum sliding window sizes.