Balancing robustness and redundancy in the design of environmental sensor networks

This thesis proposes a new approach in the design of Environmental Sensor Networks (ESN) by achieving the highest possible robustness with minimal redundancy. The proposed methodology produces the optimal number of sensor nodes require to best achieve its purpose, determines the optimal placement of sensor nodes, and investigates the impact caused by noise or gaps in the data. Noise and sensor data gaps are usually resulting from sensor errors (e.g., biofouling, electronics noise) or communication failures. The distribution of sensor nodes in a given region is proposed using Evolutionary Algorithm (EA) as the optimisation tool. The main advantage of EA is the fact that it can test a large number of possible solutions without bias from local optima. The algorithm compares the best possible configuration of sensor nodes in an ESN using fitness function as the difference between the result yielded from the network and the historical data as a validated environmental models. The results obtained were promising, however, the proposed methodology relies on historical data. To overcome this limitation a set of mobile platforms (e.g., drones, animal-carrying sensors, robots, boats of opportunity) is simulated as collecting data from the environment (i.e., from a large modelling output set). The results of the mobile platform readings are then spatial-temporally interpolated and the results used by the EA to propose a configuration of the first ESN. Validation for the proposed methods in this thesis is achieved by formulating and running the methods in a form of simulation study. The effectiveness of each ESN design produced in representing the RoI is compared against SouthEsk data model (i.e., as a representation for the actual measured value in the RoI). The performance of the proposed methods is also compared with some other methods in ESN design, including with expert knowledge. The main contributions of this work are the measure of ESN representativeness which enable to assess and to compare the performance of different ESN designs; the method to find optimum ESN design which best represents a region of interest with a balance between minimum redundancy and maximum robustness; and the use of mobile platforms for data sampling to capture environmental behaviour in a region of interest which is useful for ESN design in the absence of historical data.