The purpose of this paper is to present a new methodology in detecting novelty in sensor networks, which is based on an application of neural networks. Sensor networks are anticipated to become the main data acquisition technology of the near future, bringing up a wealth of information about the objects under measurements and environment. However, they may become the object of malicious actions and attacks. Novelty can result from a change in a sensor's environment, malicious activity, or sensor malfunction. With applications ranging from robotics to security, novelty detection is an important problem.
In our group we are trying to investigate a possibility of developing novel contents based sensor networks protocols, enhancing accuracy, reliability and security of sensor networks. Efficiency improvement is based on application of association models, which correlate information streams acquired from sensors located in close proximity to each other. We are developing the sensor networks anomaly detection system, which should incorporate a variety of intelligent agents deriving different models and applying them for enhancing various sensor networks performance indicators.
The presented implementation is build upon a time based multilayer perceptron. It presents a new function prediction neural network model that learns and runs noticeably faster than the classical time-based multi-layer perceptron. This network excels at periodic function prediction, a problem that has been traditionally considered very difficult to solve. It uses prediction to determine novelty, in effect computing a confidence level for each sensor in the network. To this end, an application has been created that allows any set of sensors to be used as inputs and any topology of time-based multi-layer perceptron or modified time-based multi-layer perceptron to be used in prediction. A set of experiments to test the implementation and investigate the parameter choice and relationship are explained and the results thereof are presented. The final section presents analyses of the results and the discovered limitations of the system.
Colloquia Series page.