Spatiotemporal analysis of demand in stochastic locational planning problems through the application of an artificial neural network

Abstract

The aim of this paper is to analyze and predict the spatio-temporal point pattern of demand in stochastic locational planning problems. Demand is not considered static but spatially and temporally in flux. As a result, the potential solutions are not clearly defined. The proposed methodology, handles stochastic locational planning problems in three levels providing, diachronic analysis of spatial point patterns of demand, forecasting their future evolution and finally locating p facilities in an optimal way. More specifically, the spatial point pattern of demand is analysed over time in a nearest neighbour context. Secondly, the approach provides the ability to predict, through the utilization of neural networks, how the pattern of demand will evolve. Finally, it locates supplying centres and allocates demand to them by utilization of fuzzy logic technique.

The application deals with the location of fire brigade vehicles in the metropolitan area of Athens. Demand prediction, based on diachronic data, allows vehicles to be placed in locations that will minimize distances from expected incidents. Diachronic data are analyzed by the neural network interpreting the spatiotemporal pattern of demand, thereby predicting the possible location of future events. The location of mobile units and the allocation of dem and to them is then carried out by means of fuzzy logic. Finally, in an attempt to assess the effectineness of the proposed methodology, the results, both in terms of demand and supply, are compared to the equivalent emergency calls and the proposed locations under the p-median formulation.