5. Conclusions

A deep belief net (DBN) composed by multiple restricted Boltzmann machines (RBMs) and multilayer perceptron (MLP) for time series forecasting were introduced in this chapter. The training method of DBN is also discussed as well as a reinforcement learning (RL) method; stochastic gradient ascent (SGA) showed its priority to the conventional error back-propagation (BP) learning method. The robustness of SGA comes from the utilization of relaxed prediction error during the

Figure 9.

50

Prediction results by DBN with BP and SGA. (a) Prediction result of CO2 data. (b) Prediction result of Sea

level pressure data. (c) Prediction result of Sun spot number data.

Time Series Analysis - Data, Methods, and Applications

benchmark CATS data, and real time series datasets showed the effectiveness of the DBN. As for the future work, there are still some problems that need to be solved such as how to design the variable learning rate and reward which influence the learning performance strongly and how to prevent the explosion of characteristic

Series Total data Testing data DBN with BP DBN with SGA Lynx 114 14 19-16-1 7-14-2 Sunspots 288 35 20-18-11-1 10-12-12-17-2 River flow 600 100 20-17-18-1 19-20-5-18-5-2 Vehicles 252 52 20-13-20-1 20-11-5-2 RGNP 85 15 18-20-1 19-15-2 Wine 187 55 16-15-12-1 18-12-13-11-2 Airline 144 12 15-4-1 13-7-2

Data DBN with BP DBN with SGA Lynx 0.6547 0.3593 Sunspots 999.54 904.35 River flow 24262.24 16980.46 Vehicles 6.0670 6.1919 RGNP 771.79 469.72 Wine 138743.80 224432.02 Airline 380.60 375.25

Training Deep Neural Networks with Reinforcement Learning for Time Series Forecasting

© 2019 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/ by/3.0), which permits unrestricted use, distribution, and reproduction in any medium,

, Masanao Obayashi<sup>1</sup>

, Shingo Mabu1

\*, Takaomi Hirata<sup>1</sup>

\*Address all correspondence to: wu@yamaguchi-u.ac.jp

eligibility trace in SGA.

Author details

53

Table 5.

Table 6.

Prediction MSE of time series data of TSDL.

DOI: http://dx.doi.org/10.5772/intechopen.85457

Takashi Kuremoto<sup>1</sup>

and Kunikazu Kobayashi<sup>2</sup>

1 Yamaguchi University, Ube, Japan

2 Aichi Prefectural University, Nagakute, Japan

Size of time series data and structure of prediction network.

provided the original work is properly cited.

#### Figure 11.

Prediction results of natural phenomenon time series data of TSDL. (a) Prediction result of Lynx; (b) prediction result of sunspots; (c) prediction result of river flow; (d) prediction result of vehicles; (e) prediction result of RGNP; (f) prediction result of wine; and (g) prediction result of airline.

learning process, i.e., different from the BP method which adopts all errors of every sample to modify the model. Additionally, the optimization of the structure of DBN was realized by random search method. Time series forecasting experiments used

Training Deep Neural Networks with Reinforcement Learning for Time Series Forecasting DOI: http://dx.doi.org/10.5772/intechopen.85457

