Adaptive Sampling in Wireless Sensor Network - Article Example

WIRELESS SENSING SYSTEMS ADAPTIVE SAMPLING STRATEGY: CASE STUDY OF HOUSEHOLD WATER CONSUMPTION. Wireless sensing systems (WSS) are currentlygaining wide use in data collection in various applications such as environment monitoring, energy consumption monitoring and control, and industrial condition monitoring. The WSS systems are predominantly battery driven with low data rates; therefore, it is undesirable to gather all available data without considering the dynamics of the monitored environments or processes. This study investigates an adaptive sampling strategy for WSS aimed at reducing the number of data samples by sensing data only when a significant change in these processes is detected. This detection strategy is based on an extension to Holts Method and statistical model. To investigate this strategy, the water consumption in a household is used as a case study. A number of performance metrics are used to evaluate the adaptive sampling strategy, including sampling fraction, missing ratio and sampling performance. The experimental results show that the proposed strategy over-performs compared to two existing sampling algorithms. Keywords: adaptive sampling algorithm, exponential double smoothing, time series forecast, and Wireless Sensing Systems 1. Introduction Wireless sensing systems has recently become a topic of interest to many researchers, especially in terms of in-network data collection. Wireless sensor networks (WSN) are considered one of the reliable environmental monitoring systems that are constructed using various techniques and numerous algorithms, which are studied and analysed according to the deployment plan. Courtesy of its targeted sensing accuracy, cost reduction, and energy consumption, the technology has been used in studies and experiments that have been conducted and implemented with great precision. Although many researchers are continuously working in this field, it appears less than satisfactory for the rapid development and technological improvements of the future. For example, for event detection; adaptive sampling is used to balance the sample rate when an abnormal behaviour or event in the collected data is detected. *Author to whom all correspondence should be addressed: Email: N.xxxx @xxx.ac.uk; phone: +44(0)153xxx; fax: +44(0)150xxx2. This paper addresses the problem of capturing the essential details from the measurement of household water temperature while minimising the energy consumption of the sensor’s battery. The existing adaptive sampling techniques proposed for data acquisition have critical limitations; particularly the concern of energy consumption. Hence, we propose an adaptive sampling algorithm based on time series statistics and the concept of TCP Reno congestion control. Our approach is designed to generate data from the source node only when a significant change is detected. Gupta et al. (2011), carried out research work on monitoring pollution levels from car exhaust gases. They implemented their work in a novel sampling algorithm for the design and evaluation of the datasets. This team used the EDSAS (Exponential Double Smoothing based Adaptive Sampling) algorithm, which is considered a time series technique. EDSAS was deployed as a data reduction technique based on predictions for an irregular sampled time series known as Wright’s Extension to Holt’s Method. It also incorporates the use of EWMA (Exponential Weighted Moving Average). Furthermore, Le Borgne et al. (2007) proposed an adaptive model selection technique for time series prediction. The technique is considered a lightweight, online algorithm for a temperature time series acquired from a sensor deployed in the real world. It was implemented to adaptively choose the best performing model for satisfying data prediction algorithm (Le Borgne et al., 2007). The selected model called Algorithm Model Selection (AMS) was based on an Auto Regression (AR) model, whose parameters can be updated in real-time. The authors stated that the prediction model allows the detection of outliers to be recovered with more possible testing values. However, other authors (de Aquino et al. 2007; Huang et al. 2011) concentrate on using different mechanisms for the purposes of monitoring increases in network lifetime and decreasing delays and energy consumption. Data stream and prediction filtering schemes for sampling are used respectively to solve the energy wastage problem. An adaptive data collection strategy was designed to allocate a number of updates that were allowed to be directed to the sink. The lifetime-constrained strategy proposed was successful. However, losing data was recorded as having a more negative influence on data accuracy (Tang & Xu, 2008). Other researchers have proposed adaptive sampling techniques that focus on the reduction of the resources used by WSNs in a decentralised manner. The technique proposed by Bhuiyan and Wang (2013) and Jain and Chang (2004), involves new samples to be collected. Notably, when the sampling rate violates a set range, a new one is requested from the server. Masoum et al. (2013), targeted applications that can tolerate changes in sensor values as long as measurements fall outside a specific range. Cheng et al. (2010), also presents another adaptive sampling approach that is focused on the efficiency of energy consumption. Their work proposed an approach using a new matrix called EDCA (Efficient Data Collection Approach) to lower the sampling rate and make sure fewer packets are transmitted. Moreover, Arici and Altunbasak (2004) proposed an adaptive sensing method that offers a compromise between system lifetime and distortion of the reconstructed image. The main idea in their method is to keep the sensors more active at intervals when the measurements show small variations. Zhou and De Roure (2007) proposed yet another sampling technique that is applied in flood warning systems. Jin et al. (2010), designed and proposed a water environment monitoring system that was based on WSNs. This system can easily be configured as a random constraint or parameter monitoring network. Additionally, another system of remote water quality detection measuring and monitoring based on WSNs, and a specified function called Code Division Multiple Access (CDMA) technology, has been proposed by Wang et al. (2010). It was designed mainly for the purposes of reducing energy consumption and ensuring effective information acquisition in WSNs. Most of the above water applications based on WSNs were designed to detect the quality of water or water pollution. This paper presents part of the research outcomes from an ongoing EU funded project Integrated Support System for Efficient Water Usage and resources management (ISS-EWATUS). The project receives its funding from the European Unions Seventh Framework Programme for research; technological development and demonstration under grant agreement no 619228. The project aims to develop several innovative ICT methods for the fulfilment of water-saving potential in the EU (http://issewatus.eu).We are aiming for the best possible energy saving algorithm, as well as the accuracy of temperature readings. The experiment was conducted using a household’s water temperature as a case study. The paper is structured into five sections as follows: Section 1 provides an introduction into wireless sensing system algorithms. Section 2 presents the prediction model to analyse the raw data collected. Section 3 explains the details of our research methodology, TCP principle along with the proposed adaptive sampling algorithm. Section 4 presents the results and discussion of the outcomes of the techniques implemented. The conclusion and suggestions for future work can be found in section 5, which concludes the paper. 2. Adaptive Sampling Algorithm: Prediction Model Details The prediction model of adaptive sampling TCP is based on a prediction method called Exponential Double Smoothing (EDS) method. EDS forecasting formula results were gathered using smoothing parameter α as the highest for the previous observation or last reading , and at a decreasing weight (1-α) to the distant observation, while is the new step estimate and is the current Exponential Weighted Moving Average (EWMA) using recursive equation (2-1) (Gupta and Shum , 2011). = (2-1) , t >0 Equation (2-1) can be written as equation (2-2) considering the forecast error for estimated period t, where is the new reading estimate and is the current reading. = (2-2) At this stage, the EDS forecast method was chosen because it was found to forecast results accurately when adopted with trends or changes in the sampling interval. We used equations (2-3) and (2-4) for calculating the EDS model. Therefore, adjusting the smoothing estimate from the trend and the previous period estimate in equation 4. Where represents the trend at interval t, is the smoothing parameter used for the linear trend, and then equation (2-4) updates the trend as the difference between the last two estimates; this method, which uses trends, is known as Holt’s Method. Holt’s Method was chosen in our analysis, along with the two parameters, to estimate the expected increase or decrease in k step size, is given in equation (2-5) (Wright, 1986). =+) (2-3) (2-4) ≥ 1 , t >0 (2-5) Our proposed algorithm works when the sample rate reaches the maximum step size () of a defined value. The data continues to be sampled until it reaches the value specified. It then stays at that interval, provided that the forecast error for period t () stays below error tolerance (δ). On the other hand, if the sample interval reaches, then the sample interval stays at and the algorithm will check for changes; if an event is detected then the sample interval will drop to half the current sample interval. However, if the sample interval is less than, the next sample interval is double the current. For the last stage, adjustment feedback was necessary to maintain data fidelity and minimise the number of false misses. False misses are defined as important changes/events that happen in the environment but cannot be detected when sampling intervals are at the maximum step size. For the adjustment feedback, we used EWMA as the basis of a change detection mechanism. In addition, we needed to use many other parameters such as alpha () used for the smoothed data series (0 Read More