Data Attribute-based Prediction Models Improvement Research In WSN

With the development of sensor, computer and wireless communication, wireless sensor network (WSN) becomes one of the hottest research fields. This paper introduces WSN, its characteristics, and future development of WSN and wireless sensor according to the characteristics. Because WSN has the characteristics of numerous sensors, high time relevance and spatial relevance and limited resource in single sensor, data fusion technique can reduce significantly the communication for sensors and power consumption, prolong the lifetime of wireless sensors, and improve the efficiency of sampling data, and accuracy of information delivered. There is a lot of research about data fusion, most of which focus on application layer and network layer. This paper introduces the nowadays data fusion techniques, including tree-based data fusion、spatial-temporal data fusion, route-driven data fusion, distributed compression data fusion and prediction-based data fusion, and details the work process of prediction-based data fusion and three kinds of prediction-based data fusion: probabilistic model, autoregressive model and Kalman filter.Nowadays most researches in prediction models are combination of several prediction techniques, and the prediction model will not change until its performance is very bad, so this paper proposes to adapt prediction models according to the specific environment in time. Here we analyze the time series of temperature, and find out that the weather can influence the time series seriously, and every temperature datum has the corresponding weather, so the paper defines the influence of environment over data as data attribute, and introduces a method how to calculate the data attribute and update it. Here we add a data attribute item to count the influence in the prediction models, that the sensitiveness of prediction models to the environment is improved. In order to prove the important effect of data attribute in performance of prediction models such as autoregressive model and Kalman filtering model, in this paper the related parameters (updating factor and weight factor of data attribute) are set in a given rang, and simulation experiments are conducted basing on the temperature data. The result shows that the data attribute can actually improve the prediction performance in update rate and prediction error. In order to reduce the update rate which is very important in prediction performance, this paper also introduces the genetic algorithm in optimizing the related parameters, and the experiment result shows that the prediction models surely get further improvement.