Research On Proactive Data Service Model And Method For Sensor Stream Processing

The sensor stream refers to a kind of sequence data from devices and sensors.It can reflect the state of the physical world and has the characteristics of continuous,time-varying,complex,and uncertainty.It is a hot topic to improve the value density and sharing degree based on service abstraction and data fusion.Existing methods are faced with many challenging problems including how to capture changes of sensor stream sources and dynamic data correlations for adjusting data mapping strategy adaptively,how to proactively and effectively capture valuable information form sensor streams in a data-driven way,how to guarantee the timeliness requirement of concurrent or dynamic data requests.This paper focuses on these difficult problems,and the main contributions are as follows:1.This paper proposes a self-adaptive proactive data service model.This paper proposes a Self-Adaptive Proactive Data Service(APDS)model for servitization modeling,time-varying sensor streams,and data correlations.The APDS model integrates an adaptive controller based on a “decision-execution-learning-adjustment” adaptive loop,which can adapt to the dynamic stream sources and proactively select cooperative services.This paper proposes a learning-based heuristic strategy execution algorithm to ensure service collaboration effectiveness with a limitation of computing resources.The experimental results show that the APDS model supports dynamic updating and processing of stream sources at runtime,which can significantly reduce the system load of the services(average 30% CPU load and 26% memory load)while ensuring the effectiveness of event capturing,and can improve the efficiency of service collaboration by reacting with up to 67% more events from other services.2.This paper proposes a data-driven continuous optimization method for proactive data service instances.This paper considers two kinds of data correlations and proposes a Dynamic Time Wrapping(DTW)-based correlation analysis algorithm and an FP-based Maximum Frequent item Sets Mining(FP-MFSM)algorithm for the uncertainty and complexity of sensor streams.This paper proposes a continuous optimization method for proactive data service instance based on period data associations and service's effect,which enables service instance to access and analyze related sensor streams,and to evolve at runtime.The experimental results show that optimized service instances can capture more accurate and richer events(increase the precision by 22%,the recall rate by 9%,and the coverage of the stream sources by 25% at average),which can better support user's decision.And there is no obvious decline in effectiveness as the service keeps running.3.This paper proposes a timeliness guarantee method for proactive data service.Considering the intersection and repetition of event types required from different applications,this paper proposed an event matching(MI-Tree)algorithm based on a multi-level indexing mechanism and matching tree for concurrent and dynamic application requests and the time-sensitive requirement of applications.The algorithm avoids performing repeated processing operations to optimize the execution of proactive data service.Experiment results show that the optimized proactive data service can better meet the requirement of timeliness of applications on event accessing,the service response time remains within 20 ms with the number of concurrent requests increases,and it can cope with added or canceled application requests at any time.The research results have been applied experimentally in actual applications from the power grid,and the experimental results further confirm the effectiveness of proposed methods and mechanisms.