The new feasibility analysis model proposed in this study for coastal projects consists of three interrelated decision support models: 1) Artificial Neural Network (ANN) to determine the rates and capacity of cargo by considering the economical development of hinterland 2) Queuing model to determine the waiting to service time and the berth occupancy ratios by waiting time modeling of ships using discrete queuing simulation 3) Importance Sampling Monte Carlo (ISMC) to simulate ship arrivals/departures from the quays and to estimate income/expenditure parameters of the coastal project. As a case study, the proposed model was applied to the Iskenderun Pier in Turkey and the future loading/unloading cargo rates of pier were predicted by ANN's. The superiorities of this proposed simulation-ANN model to other classical investment planning methods were the inclusion of uncertainties in the investment parameters like the change of cargo and costs variables in time, and the determination of project benefit/cost with an improved accuracy when compared to classical decision support models.
TopDecision Support Model
The ANN sub-model predicted the annual shipment rates in the Iskenderun Bay between the years of 2015 and 2030. This model was interrelated with ISMC simulations, since the maritime trade in the bay was mainly affected in the last decades by the successive wars of Syria and Iraq; therefore, a large extent of uncertainty was inherent in its future predictions. The neural network is a multivariate, nonlinear, nonparametric artificial intelligence system, which consists of a large number of interrelated processing units (neurons) to simulate the human brain learning. ANN’s are non-linear prediction models of natural systems, which can be used as alternatives to regressive models in coastal projects (Balas & Koc, 2002; Balas & Balas, 2002; Balas et al., 2004). In recent years, neuronets have been successfully used for analysis of coastal environments, such as for time series processing (Deo & Kumar, 2000), tidal level forecasting (Tsai & Lee, 1999), pattern classification (Deo & Naidu, 1999), wave data assessment (Tsai et al., 1999; Balas et al., 2004) and structural failure predictions (Mase et al., 1995).
The artificial neural network (ANN) is a non-linear robust prediction method that can satisfactorily handle the randomness and uncertainty inherent in data sets and complex natural systems. Neuronets exhibit the characteristics of “biological” networks to simulate the human brain learning and they have been successfully used for the prediction of environmental processes. In summary, artificial neural networks can demonstrate the learning and adaptation capabilities of biological neural systems by predicting stability numbers from experimental data sets considering the change in environmental conditions. On the other hand, fuzzy systems can effectively utilize the uncertainties inherent in human knowledge, but they do not have the capability of learning. Therefore, neural networks have more “generalization ability” (functional approximation capability) than the fuzzy systems in using a database. However if the database is limited and contains qualitative information, fuzzy systems can be alternatively used as design models. The advantage of fuzzy systems is that, they can handle human based information such as experience and judgment, and can consider qualitative data described by language.
Neural networks are simplified models of the human brain. Connections between neurons are axons and dendrites, connection weights represent synapses, and the threshold approximates the activity in the soma. The response of network:
,where y
j is the output of the j
th neuron, x
i is the i
th input signal, w
i is the connection weight from the i
th input neuron to j
th neuron, t
j is the threshold or bias value of the j
th neuron and f(x), is a non-linear function.