以大數據分析建立船殼退化狀況的預測模型

ESTABLISHING A PREDICTION MODEL OF HULL DEGRADATION USING BIG DATA ANALYSIS

本研究透過統計分析工具分析六艘船舶的航行大數據資料，找尋與主機燃油質量流率相關之參數作為反應變數，建立船殼退化狀況之預測模型，與實際狀況相互驗證評估船舶維修效果，並且利用存活曲線找出船殼退化的時間點推估維修週期，藉此降低船舶維修對航運之影響，同時兼顧燃料成本問題。將同一艘船舶當中以轉速為零的時間區分航段，由於航行於沿岸時易受風、浪、洋流的影響，因此選擇以航行於亞洲與美洲之間的航段建立模型，此定義為遠洋航段。以時間序列轉換函數模型（Regression with ARIMA error）的方式，將一個遠洋航段的反應變數代入主成分分析（Principal Component Analysis）建立模型，並將依時間排序的航程代入預測模型，比較預測值與實際值之間的差距，最後，利用平均絕對值誤差百分比（Mean Absolute Percentage Error）繪製船殼退化監控圖，從MAPE的變化趨勢觀察船殼的退化狀況以及維修效果，並找出船舶可能需要維修的時間點。

"The study aims to analyze six vessels by statistical tools and find the relationship of main engine mass flow rate with relevant variables. According to the model of fitting fuel consumption, we compare to predicted values and real values, and look for the time points of the hull deterioration and infer the maintenance period. It makes that people can notice the hull status and improve the energy efficiency in the voyage. Thus, this model not only reduces the impact on shipping but considers the fuel cost. In the same ship, segments with similar voyages are grouped into one group for modeling. The shipping variables can be transformed by principal component analysis and fitted predicted model by the regression with auto-regressive integrated moving average error. In similar voyages, we predict the mass flow rate for other voyages and compute the mean absolute percentage error in the real values of the same voyage. It detects whether the hull is degraded, and the maintenance is useful or not. As a result, we can find the necessary maintenance from the monitoring charts. Besides, we can give suggestions for the maintenance period."