Minimizing Cost of Redundant Sensor-Systems with an Artificial Immune Algorithm Q. M. Jonathan Wu National Research Council Canada 3250 East Mall, Vancouver, B. C. Canada V6T 1W5 Jonathan.Wu@nrc.ca Guiliang Yin Electrical Engineering Department Yashan University, 066004, China guiliang.yin@nrc.ca Abstract – Using human immune system mechanisms, an artificial immune algorithm used for minimizing cost of redundant multi-sensor systems is presented in this paper. The search surface for finding the minimal cost of the sensor system while insuring system dependability has many local minima. The heuristic methods must be used to solve the problem. The artificial immune algorithm uses hyper-mutation to search local areas and receptor editing to escape from local minima. It is more suitable than simulated annealing and genetic algorithms for solving this problem. The results produced by the method are compared with results of simulated annealing and genetic algorithm. Keywords: Sensor fusion, dependability, artificial immune algorithm, optimum. 1 Introduction Multi-sensor fusion and integration refers to the synergistic combination of sensory data from multiple sensors to provide more reliable and accurate information. The potential advantages of multi-sensors fusion and integration are redundancy, complement-arity, timeliness, and cost of the information. Redundant multi-sensor systems achieve fault tolerance by multiple inaccurate sensors. Feasibility requires attention be paid to both dependability bounds and cost. Success in designing redundant systems depends on making the best possible trade-off at least cost. In [1]-[4], dependability measure methods of redundant systems are given in detail. After deriving the dependability expressions, an algorithm called exhaustive search guaranteed to find the minimal cost configuration using the item costs of each sensor is presented. But this algorithm is computationally expensive; it is just suitable for small-scale problems. For lager-scale problems, Tabu search, simulated annealing (SA) and genetic algorithms (GAs) are applied for finding near-optimal combination. Tabu search is often used as an alternative to simulated annealing and it has no clear stopping criteria. Tabu search, simulated annealing and genetic algorithms are relatively insensitive to the presence of local minima in the search space. With simulated annealing and genetic algorithms this insensitivity is partially obtained by the creative application of non-determinism. This non- determinism also means that the quality of the answers found by the algorithm will vary from case to case. GAs are sensitive to the reproduction strategy chosen, including mutation rates and how elements are chosen for crossover. SA is sensitive to the cooling schedule, which includes the initial temperature and the rate of decrease of temperature. The quality of the answers found and the amount of time needed to find reasonable answers are directly dependent on the reproduction strategy of a GA and the cooling schedule of an SA approach. Both the reproduction strategy and cooling schedule must be found through a process of trial and error. For neither is there a guarantee that a particular strategy nor schedule will be appropriate for all cases encountered. The immune system, with its cell diversity and variety of information processing mechanisms, is a cognitive system of complexity comparable to the brain. Interest in studying the immune system has been increasing over the last few years. Several optimal methods using artificial immune mechanism are presented recently [5-11]. A comparison of optimization performances between immune algorithm and genetic algorithms is given in reference [5] and immune algorithm shows a better result than genetic algorithm for a problem having many local optima. In this paper, we present an artificial immune algorithm (IA) for the problem, based on the way the immune system’s T-cell-dependent responses. We extend the idea to adapt the problem of minimizing cost of redundant sensor systems. 2 Dependability Measure and Opti- mization Model Redundant sensor-systems achieve fault tolerance by duplication of components. It increases the ability of systems to interact with their environment by combining independent sensor readings into logical representations. Sensor integration of highly redundant systems offers these advantages: 1) Multiple inaccurate sensors can cost 1473