Time Series Prediction Method Based on Variant LSTM Recurrent Neural Network
- PDF / 1,759,023 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 111 Downloads / 224 Views
Time Series Prediction Method Based on Variant LSTM Recurrent Neural Network Jiaojiao Hu1 · Xiaofeng Wang1 Jianru Xue2
· Ying Zhang1 · Depeng Zhang1 · Meng Zhang1 ·
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Time series prediction problems are a difficult type of predictive modeling problem. In this paper, we propose a time series prediction method based on a variant long short-term memory (LSTM) recurrent neural network. In the proposed method, we firstly improve the memory module of the LSTM recurrent neural network by merging its forget gate and input gate into one update gate, and using Sigmoid layer to control information update. Using improved LSTM recurrent neural network, we develop a time series prediction model. In the proposed model, the parameter migration method is used model update to ensure the model has good predictive ability after predicting multi-step sequences. Experimental results show, compared with several typical time series prediction models, the proposed method have better performance for long-sequence data prediction. Keywords Deep learning · Time series prediction · Recurrent neural network · Variant LSTM network
1 Introduction Time series refers to a series of observations of a phenomenon arranged in chronological order. Analysis and prediction for time series can provide us with better method of decision support. The problems of time series forecasting have been paid great attention since it was proposed. The classical methods can be classified into two categories, linear models and
This work was supported by the National Natural Science Foundation of China under Grant No. 61772416; the National Major Research and Development Plan Program of China under Grant No. 2016YFB1001004; the Key Laboratory Project of the Education Department of Shaanxi Province under Grant No. 17JS098; Thirteenth Five-Year Equipment Pre-research Project No. 30503030201-02; the foundation of the State Key Laboratory of Astronautic Dynamics.
B
Xiaofeng Wang [email protected]
1
Xi’an University of Technology, Xi’an 710048, Shaanxi, People’s Republic of China
2
Institute of Artificial Intelligence and Robotics, Xi’an Jiaotong University, Xi’an 710049, Shaanxi, People’s Republic of China
123
J. Hu et al.
nonlinear models. The prediction methods of linear time series were proposed in earlier times, and mainly include AR [1], MA [2], ARMA [3] etc. These methods are simple to compute and have good predictive effect on stationary time series, but have poor predictive effect on non-stationary time series. For these non-stationary time series data, the autoregressive integrated moving average model (ARIMA) [4] were proposed. However, the development of things is mostly nonlinear in reality. Therefore, the research for time series data prediction is mainly focused on the nonlinear models. With the development of artificial intelligence technology, machine learning method has been widely applied in many fields. The prediction method based on machine learning does not need
Data Loading...
