Research On Electric Propulsion Ship Power Load Forecasting

Short-term forecasting of ship power load is the basis of electric system, the forecastaccuracy will largely affect the economy of the ship and the stability of the power system.At present, there are many short-term electric power load prediction method, and theprediction accuracy has been improved. An electric propulsion ship is chosen as a researchobject, and a support vector machine combined with other method is used to forecast itsshort-term load.The chaotic characteristic of electric propulsion ship is analyzed by Chaos Theory, atthe same time, the load time sequence of phase space reconstruction and the maximumindex calculation is completed, which validate chaos character of electric propulsion shipload. So it laid the foundation of the load forecasting.Aiming at the parameter selection of SVM, a new combinatorial optimizationalgorithm based on the iterated local search and adaptive particle swarm algorithmcompletes the parameter selection automatically. The optimization algorithm inherits theadvantages of the particle swarm algorithm including simple and optimization ability, andovercome the shortcomings including the slow convergence and easy to fall into the localoptimum. Through the various tests of functions, the advantage of the algorithm is proved.The training data set is weighted based on the pretreatment of power load time series,and the prediction model is constructed. The parameters of support vector machine arecalculated by combinatorial optimization algorithm. Then the short-term load is predictedusing the calculated parameters. The simulation results show that this method has the higherforecast precision compared with the traditional support vector machine and neural networkmethod.The electric propulsion ship power load is decomposed into a series of regular loadthrough the wavelet decomposition. Each of the load components are predicted usingsupport vector machine model based on the iterated local search and adaptive particleswarm optimization. Finally, the final prediction result is the synthesis of all loadforecasting results. Simulation results show that the method can further improve theaccuracy of load prediction.