Author(s)

M. Carnero, J. Hernández & M. Sánchez

Abstract

This paper presents a meta-heuristic approach for the optimal location of sensors in chemical plants. Within the Tabu Search framework, a Strategic Oscillation Technique around the feasibility boundary is implemented that provides an effective balance between intensification and diversification over the intermediate to long term of the search. The procedure is applied to minimize the instrumentation network cost subject to precision and availability constraints on variable estimates for linear processes. Its performance is compared to other evolutionary techniques for two case studies. Results show the algorithm succeeds in locating the optima using lower computation time. Keywords: sensor network design, stochastic optimisation, tabu search. 1 Introduction A reliable and complete knowledge of current plant state is essential for plant monitoring, regulatory and supervisory control, real time optimisation, planning and scheduling, etc. The quality and availability of variable estimates strongly depend on the sensor network installed in the process. The design of sensor networks consists in selecting the type, number, accuracy, failure rate, and location of new sensors that provide the quantity and quality of information required from the process. Frequently it is necessary to satisfy constraints only on a subset of key measured or unmeasured variable estimates. In this case a general sensor network is designed without knowing in advance the cardinality of the optimal sensor set. As only a subset of variables

Keywords

sensor network design, stochastic optimisation, tabu search.