Cyber-Physical System (CPS) becomes more and more importance from industrial application (e.g., aircraft control, automation management) to societal challenges (e.g. health caring, environment monitoring). It has traditionally been designed to one speciﬁcappli cation domain and to be managed by a single entity, implemented communication between physical world and computational world. However, it still just work within its domain, and not be interoperability. How to make it into scalable? How to make it reusing? These questions become more and more necessary.
In this paper, we are trying to developing a common CPS infrastructure, let it be an innovative CPS crossing multiple domains to broad use sensors and actuators. Here, we implement a technique for automatically build a model according to the sensor data in diﬀerent domains. And based on our approach under continuous situation, it could identify the sensor values right now or estimate next few time step, which we call spatial model or temporal model.
In this section, we will describe recent CPS researches in conference publications. Many existing CPS solutions are based on stovepipe architectures. These studies include many applications in road traﬃc management, energy system, power infrastructure, and health care(monitoring devices). It basically focus on cyberizing the physical and physicalizing the cyber. The challenge of integrating computing and physical processes has been recognized for some time in cyberizing part. Time synchronization is diﬀerent implemented in software execution since there is no statement like ”current time is t” and no semantic notion of time passing.
DATA ANALYSIS AND EVALUATION
In this study we utilized a data set from government website www.weather.gov. They provide reliable sensor information from every weather station (e.g. air temperature, dew temperature, visibility, pressure, longitude, latitude and elevation), which spread over a geographical space. We apply our methodology on these multiple variables of interest to verify the eﬀectiveness of our approach. We collect 98 weather stations deployed at diﬀerent locations in New England area (contains the states of Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, and Connecticut) and New York State. It contains weather variable values reported at one hour between May 30, 2011 and June 15, 2011.
SPATIAL MODEL OF SENSOR DATA
After ﬁtting lots of diﬀerent statistical variable selection models to a given data set, we are now focus on building the model. This section details the prerequisites that how to ﬁnd best ﬁtted method after data analysis. We usually increase the degree of the polynomial till the added term is not statistically signiﬁcant. At the same time, increase the interaction part also could enhance the model and decrease the error of estimated value.
TEMPORAL MODEL OF SENSOR DATA
Since each correlation between those variables should be stable in some degree, we are trying to ﬁgure out their regularity from the correlations between each of them that we calculated by our outlier detection system. If we could discover the regularity of correlation between two variables in a ﬁxed time period, then we could easily determine the expected values over time at each spatial coordinate. We use extrapolate temporally to predict sensor values for next several time steps.
MODEL PERFORMANCE ANALYSIS AND EVALUATION
Figure 6.1 shows the outlier detection performance of our scheme. The colored dots in the ﬁgure represent the sensor readings locations in the model applied to observed pressure values. The range of pressure is from low to high marked as red, orange and yellow. The dataset is still used those 90 weather sensor readings deployed in Northeast United State.
We continue to apply our model on temperature temporal extrapolation to predict the next time slot value of sensor readings. Since each correlation between those variables should be stable in some degree, we are trying to ﬁgure out their regularity from the correlations between each of them that we calculated by our outlier detection system. If we could discover the regularity of correlation between two variables, then we could easily determine the expected values over time at each spatial coordinate.
In this paper, we have analyzed several sensors data relationship, implemented a method for spatial model and temporal model to detect sensor values no matter at diﬀerent spatial coordinates at an instant time or detect the expected value of sensors at all coordinates in next several time step. Our spatial model is modeling of diﬀerent domain sensor data based on multiple regression, and temporal model utilizes exponential smoothing to evaluate next time step model.
Our temporal model shows promise in detecting next time step sensor data values, provides great performance in evaluating next one time step model in weather sensing application, and as the same as real weather prediction model, especially under the situation that add additional sensor data, then re-do the prediction. It could illustrate that our model is not domain speciﬁc, and could be applied in any application domains with continuous sensor data of Cyber-Physical System. We do believe our research could help setting up a scalable deployment of Cyber-Physical System.
Source: University of Massachusetts Amherst
Author: Xianglong Kong