Growing-type WASD for power-activation neuronet to model and forecast monthly time series

Zhang, Yunong; Lao, Wenchao; Jin, Long; Chen, Jinhao; Liu, Jinrong

doi:10.1109/icca.2013.6564876

Cited by 9 publications

(4 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that we choose the level of the time series L t (i.e., the average value of the time series) as R t throughout the paper; and that the WASD (weights and structure determination) neuronet [11] will be exploited to estimate R t in the our further work for higher accuracy. In the rest of this section, the IHTPES, MSFD, and WCC methods are investigated for trend estimation, seasonality estimation, and global time series estimation, respectively.…”

Section: Methodology Descriptionmentioning

confidence: 99%

Time series forecasting via weighted combination of trend and seasonality respectively with linearly declining increments and multiple sine functions

Lao

Wang

Chen

et al. 2014

2014 International Joint Conference on Neural Networks (IJCNN)

Self Cite

View full text Add to dashboard Cite

In this paper, a novel weighted-combination-ofcomponents (WCC) method is proposed for modeling and forecasting trend and seasonal time series, and such a method is based on decomposition model which regards the time series as the weighted combination of trend, seasonality and other components. Specifically, the Holt's two-parameter exponential smoothing (HTPES) method is improved (for short, the IHTPES method) to evaluate the trend with linearly declining increments; and the multiple sine functions decomposition (MSFD) method is developed to evaluate the seasonality. Then the weighted combination of the evaluations is obtained to estimate the global time series. Numerical experiment results substantiate the effectiveness and superiority of the proposed WCC method in terms of modeling and forecasting time series from the NN3 competition.

show abstract

Section: Methodology Descriptionmentioning

confidence: 99%

Time series forecasting via weighted combination of trend and seasonality respectively with linearly declining increments and multiple sine functions

Lao

Wang

Chen

et al. 2014

2014 International Joint Conference on Neural Networks (IJCNN)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Because we are dealing with time‐series, an approach similar to the neuronets in Zhang et al ((2013), (2019)), which is based on the

K

‐order Taylor series, is employed. Considering that

y_{t}, y_{t - 1}, y_{t - 2}, \dots, y_{t - M}

is a time‐series, where

y_{t - 1}, y_{t - 2}, \dots, y_{t - M}

corresponds to the neuronets variables

X_{1}, X_{2}, \dots, X_{M}

, respectively, and

y_{t}

corresponds to the neuronet's output target

Y

, the neuronet model performs a nonlinear functional mapping from the past observations to the future value

y_{t}

.…”

Section: The Mi‐wasdtsn Modelmentioning

confidence: 99%

“…Furthermore, several activation functions, such as power, signum, sine and square wave, are employed on neuronets in Zhang et al ((2013), (2019)) and Zeng et al (2020). In here, our approach is based on the sigmoid function since it occurs between two points (0 to 1).…”

Section: The Mi‐wasdtsn Modelmentioning

confidence: 99%

A weights direct determination neuronet for time‐series with applications in the industrial indices of the Federal Reserve Bank of St. Louis

Mourtas

2022

Journal of Forecasting

View full text Add to dashboard Cite

The shortcomings of conventional back-propagation neuronets, such as slow training speed and local minimum, are known to be addressed by neuronets trained under the weights-and-structure-determination (WASD) algorithm. Derived from power activation feed-forward neuronets, a multi-input WASD for time-series neuronet (MI-WASDTSN) model is presented in this paper. The MI-WASDTSN is equipped with a novel WASD for time-series (WASDTS) algorithm, for handling time-series modeling and forecasting problems. Employing a power sigmoid activation function, the WASDTS algorithm handles the model fitting and validation by determining the optimal input variables number and the weights of the MI-WASDTSN. More specifically, the WASDTS algorithm finds and holds only the activation function powers that reduce the model's error during validation. Applications on Federal Reserve Bank of St. Louis (FRED) industrial indices under three different patterns of time-series validate our MI-WASDTSN model in order to demonstrate its outstanding learning and forecasting performance. In addition, to support and advance the findings of this work, we created a MATLAB repository for interested users, which is freely available via GitHub.

show abstract

“…mance of the NN. In previous researches, activation functions are almost all continuous functions, such as power functions [11], Gaussian functions [12] and Chebyshev polynomials [13]. NNs activated by these functions are shown to have the ability to achieve the approximation of nonlinear continuous functions effectively.…”

Section: Introductionmentioning

confidence: 99%

Uniqueness logic represented via decimal numbers with WASD neural network

Wang

et al. 2014

2014 10th International Conference on Natural Computation (ICNC)

Self Cite

View full text Add to dashboard Cite

A novel concept, uniqueness logic represented via decimal numbers (UL-D), is proposed and defined in this paper. Aiming at achieving the UL-D, we construct a neural network (i.e., NN) based on weights-and-structure-determination algorithm (i.e., the resultant WASD-NN). Differing from the back-propagation neural network (BP-NN) adjusting weights by lengthy iterative process and being unable to acquire the optimal structure adaptively, the WASD-NN can determine the optimal weights directly and the optimal structure automatically. Note that the UL-D is a nonlinear discontinuous mapping, of which the approximation has rarely been investigated before. In this paper, we firstly investigate the WASD-NN activated by commonly used continuous power functions, with corresponding numerical experiment results less satisfactory. By understanding the nature of the UL-D, the WASD-NN activated by discontinuous signum function is thus creatively built up, and the numerical experiment studies demonstrate well the efficient and superior approximating ability of the signum-function activated WASD-NN in achieving the UL-D.

show abstract

Growing-type WASD for power-activation neuronet to model and forecast monthly time series

Cited by 9 publications

References 15 publications

Time series forecasting via weighted combination of trend and seasonality respectively with linearly declining increments and multiple sine functions

Time series forecasting via weighted combination of trend and seasonality respectively with linearly declining increments and multiple sine functions

A weights direct determination neuronet for time‐series with applications in the industrial indices of the Federal Reserve Bank of St. Louis

Uniqueness logic represented via decimal numbers with WASD neural network

Contact Info

Product

Resources

About