06-09-2016, 10:15 AM
1453120609-paperdbpCopy.docx (Size: 54.68 KB / Downloads: 7)
Abstract—Wireless Sensor Network(WSNs) is a network that consist of several sensor nodes which are resource constrained. These sensor node are used to gather data from external environments and send information that contains temperature data, humidity data to the sink node , this causes high energy consumption. This problem can be overcomed by data prediction method. Prediction of data is a technique used to save energy in WSNs there by reducing the amount of data traffic. In this paper we have analyzed the different data prediction algorithm and conclud with one less complex algorithm with least RMSE to be used for WSN. The algorithm considered for analysis are Derivative Based Prediction algorithm ,Auto Regressive Prediction algorithm, Moving Average Prediction algorithm and Auto Regressive Integrated Moving Average algorithm.
I.INTRODUCTION
A Wireless sensor network are spatially distributed sensors that is
having freedom to govern itself or control its own affairs [1]. These
sensors senses the environmental condition and send the sensed
data to main location through the network.The main location having
the sensor node collect all the sets of data for further processing .
These sensor node consumes more power for sensing ,communicating
and data processing. These process causes high power consumption.
Since the wireless sensor node is kept in a location which is hard to reach ,
so it is inconvient to change the battery regularly.This high power
consumption problem can be reduced by reducing the amount of data sent to
the sink node.To overcome these problems different types of data reduction
techniques have been developed. Data compression, data prediction
and inter networking process are the different types of techniques developed
to reduce the data sent to the sink node.To reduce the amount of data recieved
in the source node data compression technique is applied. This data compression
technique is involved when most recent measurement is not needed by the wsn
application. To transform the large amount of data collected into less detailed
refined data ,data aggregation method is carried out by inter networking technique
in the route where data transferring is done to the sink node. Algorthamic
approaches,stochastic process and time series forecasting are the three
subclasses of data prediction techniques. Predicting the data with least RMSE
is an approach to reduce the high power consumption without compromising data quality[2].
Predicting the data reduces the number of transmission to the sink node.In this paper we investgate and analyse
the different data prediction algorithm for wireless sensor network.
The rest of the paper contain the following : the next section gives a description of the four data prediction algorithm –DBP,AR,MA,ARIMA .Chapter four covers the simulation results. Finaly the paper is concluded in the last chapter.
II.DATA PREDICTION ALGORITHMS
A)DERIVATIVE BASED PREDICTION ALGORITHM:
The Derivative based prediction model is calculated based on a learning window.The learning window, contains two edge points and data points. The two edge points is represented as l and the data points are represented as m. The average values of the edgepoints in the learning window are connected using the slope δ.Derivative based prediction name is given to the model since the computation is similar to a computation of a derivative. Derivative Based Prediction is very efficient computationally by making use of only two edge points.
The algorithm used in dbp is explained in the following , first `m’ consecutive values (the learning window) is taken . Next `l ‘values on right edge and `l’ values on left edge (typically l should be <= m/2) is taken .Subtract average of values from right side edge from the average of those on the left side edge. Divide the resulting number by (m-l) and call it as Derivative . Add the Derivative in the last value of the learning window (the most right side value) to get prediction for m+1 th value. The (m+2) th value can be obtained by adding the prediction of (m+1)th value to Derivative and so on . For each new value we need to check if the prediction value falls within error tolerance i.e., Is my predicted value within some (e.g. 5) percentage of actual sensed value. If it is within error tolerance, it is fine and can continue adding Dervative to this current predicted value get the next value. In case predicted value is way too off, we need to regenerate the model i.e., use last m actually sensed values to calculate new derivative using process explained in the before points.