Simulation of oil spill using ANN and CA models

In this paper, the artificial neural network (ANN) used to obtain transition rules in oil spill CA model. Model parameters are difficult to obtain in many traditional oil spill models, as they cannot meet the requirements of rapid response for oil spills. Therefore, a new oil spill model - ANN oil spill CA model was established in this paper. This model can simulate the change process of oil spill by setting initial image, verification image, and impact factors. Experimental results show that the simulation results have a good performance with overall accuracy of 96.6% and Kappa coefficient of 0.826. It was also found that the consistency of simulation results is proportional to the ratio of training sample. However, the higher the ratio of the training sample, the more computation is need in the ANN training. We also studied the effect of neurons number in the hidden layer. Studies show that the consistency of simulation results becomes better with the increase of neurons number in the initial stage for good fitting rate of training sample. However, the consistency of simulation results get worse for over-fitting of training sample in following stage.