Optimisation of Long-Term Industrial Planning

Doktorsavhandling, 2006

In this thesis, long-term optimisation methods for industrial transition processes have been developed, taking monetary and environmental considerations into account. Two different methods for investment optimisation have been developed. First, an optimisation method comprising simultaneous calculation of the long-term investment strategy and the short-term utilisation scheme for a deterministic demand was developed. The method has been applied to the case of finding an investment strategy for minimising the production cost for a single hydrogen refuelling station. The problem was shown to be convex; thus the resulting solution is the global optimum. Second, an investment optimisation method using stochastic demand scenarios and multi-objective optimal control to produce the Pareto front of the two conflicting objectives \emph{expected production cost} and \emph{expected unsatisfied demand} was developed. This method was applied to the case of finding the optimal investment strategy for a combined hydrogen and hythane refuelling station. Depending on the preferences of the decision-maker, many different feasible solutions can be found. However, it was also found that, due to the uncertainty of the stochastic demand function, satisfying all the estimated demands would require a production capacity well above the mean demand, which would be very costly to maintain. In addition to the two methods for investment optimisation, a modelling approach for systems combining economic and environmental aspects has been developed as well. This approach has been used for modelling cement production facilities, taking both economic and environmental issues into consideration. In order to deal with prediction uncertainties, time series prediction using genetic algorithms was investigated as well. Discrete-time prediction networks, a novel type of recurrent neural networks, were introduced, and were shown to provide one-step macro-economic time series prediction with greater accuracy than several other methods.