Research On Small Oil Spill Model At Sea Based On BP Neural Network Optimized By Genetic And Particle Swarm Optimization

The increasing development of marine oil transportation and exploration has also brought about the problem of marine oil spill.Marine oil spill will seriously damage the marine ecological balance,so it is particularly urgent to study the emergency treatment and risk assessment technology of oil spill pollution.The numerical model of oil spill can be used to predict the diffusion behavior of oil spill,which is an important tool for risk assessment during oil spill accidents.The existing oil spill numerical models are mainly for large oil spills,and there are few studies on small oil spills.Therefore,this thesis optimizes the existing numerical model to obtain a new numerical model for the prediction of small oil spills.Different oil products（heavy oil,light oil and gasoline）were used to simulate the small oil spill diffusion simulation experiment under different environmental conditions（wind and wave）under different oil spill amounts.The oil spill image was obtained during the oil spill simulation experiment.The image was identified by genetic algorithm（GA）and edge-based active contour model respectively,and the optimal image recognition model was selected.BP neural network（BP-ANN）,genetic algorithm optimized BP neural network（GA-BP）and particle swarm optimization BP neural network（PSO-BP）were used to predict the oil spill area,and the prediction model with high precision and strong stability was selected.On this basis,a small oil spill numerical model combining image recognition and data prediction was constructed.The main findings are as follows.（1）The results showed that the infrared thermal imager was not suitable for the oil spill image acquisition under the simulation experiment,so the ordinary optical camera was used in the simulation experiment to obtain the oil spill image.The image was identified by genetic algorithm（GA）and edge-based active contour model respectively.The recognition results show that the genetic algorithm is more accurate for the image recognition of the simulation experiment.Therefore,the genetic algorithm is applied to the image recognition of the oil spill in the small oil spill simulation experiment.（2）In the oil spill simulation experiment,in the presence of wind and different wind directions,different oil and oil volume,the comparison results of three different algorithm prediction models show that the PSO-BP model has the fastest convergence speed and the highest stability,which can achieve the goal of stable prediction.The prediction performance evaluation data of PSO-BP model(R²=1,MSE=3.58e^-9～8.87e^-8)show that PSO-BP model can provide a reliable method for the prediction of small oil spill area.（3）The results of small oil spill simulation experiments under different wind and wave conditions show that the prediction and evaluation data of PSO-BP model are MSE=4.12e^-13～3.03e^-6and MSE=9.13e^-9～1.32e^-8,respectively,which have better prediction accuracy and stability than the other two algorithms.Therefore,genetic algorithm image recognition and PSO-BP model can provide theoretical basis and technical support for reliable prediction of small oil spill.