전기추진 선박의 소비 전력 예측 방법 비교 연구

Abstract

In recent years, concern for the energy efficiency of vessels has increased due to the influence of environmental pollution. Electric propulsion systems have more energy flexibility than mechanical propulsion systems. So, it can improve power efficiency by using batteries. For the utilization of batteries, it is necessary to predict the power consumption of electric propulsion vessels. In addition, researchers are studying multivariate time-series data for prediction. In contrast preceded studies were non applicability to electric propulsion systems for power consumption predicting. Because, limitation of accessible to vessel's data and the previous prediction researches were considerably studied to small ranged electrical data. According to these reasons, The research of models that capable a wide range of vessel load data is essential. In this paper, aims to predict the electricity consumption of a vessel, using real vessel data and convert to electric propulsion vessel data, and select variables that affecting vessel's electricity consumption using heuristic. The converted data includes missing values, this can cause of weakens model's accuracy, therefore multiple imputation algorithm was used for cover it. After data preprocessing, several models are created to predict time-series data. This consists of single models for comparison criteria : LSTM(Long Short-term Memory models), CNN(Convolutional Neural Network), ANN(Artificial Neural Network), DNN(Deep Neural Network), bidirectional LSTM, and conjunction models : CNN-LSTM (direct), CNN-bidirectional LSTM (direct), CNN-LSTM (parallel), CNN- bidirectional LSTM (parallel). After models creation, the experiment method was decided, considered by clear comparison. that was composed of repeat test for the model's performance validation and utilized the widely used accuracy metric : RMSE.