A quantitative model based on grey theory for sea surface temperature prediction

Abstract

In order to predict sea surface temperature (SST), combined with the genetic algorithm and the least-squares method, a GM(1,1|sin) power model prediction method based on similarity deviation is proposed. We first combined the data of two consecutive years into a new time series, analyzed the similarity of the data of the previous year, and obtained the most similar year and the corresponding new time series. Then, we established a GM(1,1|sin) power model to predict SST. In model validation, we predicted the monthly average SST from 2016 to 2020 with the data from 1985 to 2015, 2016, 2017, 2018, and 2019. The validation results showed that the maximum mean relative error (MRE) was 13.28%, the minimum MRE was 5.54%, and the average MRE and the root mean square error (RMSE) were 9.81% and 1.0627, respectively. All of evaluation metrics of Lin’s concordance correlation coefficient (LCCC) and the ratio of performance to deviation (RPD) were excellent. We iteratively predicted the monthly average SST from 2016 to 2020 with the data from 1985 to 2015, the maximum MRE was 13.91%, the minimum was 7.80%, and the average MRE, RMSE, LCCC and RPD are 11.07% 1.0603, 0.9894, and 7.497, respectively. Compared with GM(1,1), GM(1,1|sin + cos), and GM(1,1|sin) models, the proposed model outperformed these models with at least 50% in the MRE. It proves that the proposed model can be regarded as a better solution to predicting SST.

Short Term Power Load Combination Forecasting Method Based on Feature Extraction

“…Enhancing model accuracy is particularly crucial accordingly. Due to the time series and nonlinear characteristics of load sequences, numerous scholars have proposed a variety of models by simple or complex methods such as machine learning [4], exponential smoothing method [5], autoregressive integrated moving average model [6], multiple linear regression method [7], Kalman filter algorithm [8], grey prediction theory [9] and support vector machine [10]. Nevertheless, their overall forecasting accuracy has room for improvement, especially by the, challenge of applying large-scale data with universality.…”