Temperature Time Series Prediction Model Based on Time Series Decomposition and Bi-LSTM Network

Utilizing a temperature time-series prediction model to achieve good results can help us accurately sense the changes occurring in levels advance, which is important for human life. However, random fluctuations time series reduce accuracy of model. Decomposing data prior performing effectively influence and consequently improve results. In present study, we propose that combines seasonal-trend decomposition procedure based on loess (STL) method, jumps upon spectrum trend (JUST) algorithm, bidirectional long short-term memory (Bi-LSTM) network. This daily average predictions cities located China. Firstly, decompose into trend, seasonal, residual components using JUST STL algorithms. Then, determined by two methods are combined. Secondly, three original fed two-layer Bi-LSTM training purposes. Finally, achieved both merged learnable weights output as final result. The experimental show root mean square absolute errors our proposed dataset 0.2187 0.1737, respectively, less than values 4.3997 3.3349 attained model, 2.5343 1.9265 EMD-LSTM 0.9336 0.7066 STL-LSTM