自回归积分移动平均
时间序列
均方误差
自回归模型
计算机科学
系列(地层学)
组分(热力学)
数据挖掘
人工智能
算法
统计
机器学习
数学
古生物学
物理
生物
热力学
标识
DOI:10.1145/3594315.3594388
摘要
Traditional time series models have often been used to predict temperature trends. However, they may not be suitable for handling non-linear relationships in data. To address this limitation, this paper proposes a new two-step hybrid prediction model for ocean temperature forecasting, which integrates the Autoregressive Moving Average Model (ARIMA) and Long Short Term Memory Model (LSTM).The proposed model first employs the ARIMA model to capture the linear relationship in the original data. It then uses the LSTM model to correct the non-linear component of the ARIMA model's predictions, such as residuals. This approach effectively enhances the accuracy of temperature prediction. Experimental results demonstrate that the ARIMA-LSTM model outperforms both the ARIMA and LSTM models in terms of prediction accuracy. Specifically, the ARIMA-LSTM model yields lower values of normalized mean square root error and normalized mean absolute deviation. The application of this hybrid prediction model can significantly improve ocean temperature forecasting.
科研通智能强力驱动
Strongly Powered by AbleSci AI