Abstract: Time Series Forecasting (TSF) is used to predict the target variables at a future time point based on the learning from previous time points. To keep the problem tractable, learning methods use data from a fixed length window in the past as an explicit input. In this paper, we study how the performance of predictive models change as a function of different look-back window sizes and different amounts of time to predict into the future. We also consider the performance of the recent attention-based transformer models, which had good success in the image processing and natural language processing domains. In all, we compare four different deep learning methods (Recurrent Neural Network (RNN), Long Short-term Memory (LSTM), Gated Recurrent Units (GRU), and Transformer) along with a baseline method. The dataset (hourly) we used is the Beijing Air Quality Dataset from the website of University of California, Irvine (UCI), which includes a multivariate time series of many factors measured on an hourly basis for a period of 5 years (2010-14). For each model, we also report on the relationship between the performance and the look-back window sizes and the number of predicted time points into the future. Our experiments suggest that Transformer models have the best performance with the lowest Mean Absolute Errors (MAE = 14.599, 23.273) and Root Mean Square Errors (RSME = 23.573, 38.131) for most of our single-step and multi-steps predictions. The best size for the look-back window to predict 1 hour into the future appears to be one day, while 2 or 4 days perform the best to predict 3 hours into the future.
Abstract: This paper examines the forecasting performance of Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Networks (ANN) models with the published exchange rate obtained from South African Reserve Bank (SARB). ARIMA is one of the popular linear models in time series forecasting for the past decades. ARIMA and ANN models are often compared and literature revealed mixed results in terms of forecasting performance. The study used the MSE and MAE to measure the forecasting performance of the models. The empirical results obtained reveal the superiority of ARIMA model over ANN model. The findings further resolve and clarify the contradiction reported in literature over the superiority of ARIMA and ANN models.
Abstract: A model was constructed to predict the amount of
solar radiation that will make contact with the surface of the earth in
a given location an hour into the future. This project was supported
by the Southern Company to determine at what specific times during
a given day of the year solar panels could be relied upon to produce
energy in sufficient quantities. Due to their ability as universal
function approximators, an artificial neural network was used to
estimate the nonlinear pattern of solar radiation, which utilized
measurements of weather conditions collected at the Griffin, Georgia
weather station as inputs. A number of network configurations and
training strategies were utilized, though a multilayer perceptron with
a variety of hidden nodes trained with the resilient propagation
algorithm consistently yielded the most accurate predictions. In
addition, a modeled direct normal irradiance field and adjacent
weather station data were used to bolster prediction accuracy. In later
trials, the solar radiation field was preprocessed with a discrete
wavelet transform with the aim of removing noise from the
measurements. The current model provides predictions of solar
radiation with a mean square error of 0.0042, though ongoing efforts
are being made to further improve the model’s accuracy.
Abstract: Fuzzy systems have been successfully used for
exchange rate forecasting. However, fuzzy system is very confusing
and complex to be designed by an expert, as there is a large set of
parameters (fuzzy knowledge base) that must be selected, it is not a
simple task to select the appropriate fuzzy knowledge base for an
exchange rate forecasting. The researchers often look the effect of
fuzzy knowledge base on the performances of fuzzy system
forecasting. This paper proposes a genetic fuzzy predictor to forecast
the future value of daily US Dollar/Euro exchange rate time’s series.
A range of methodologies based on a set of fuzzy predictor’s which
allow the forecasting of the same time series, but with a different
fuzzy partition. Each fuzzy predictor is built from two stages, where
each stage is performed by a real genetic algorithm.
Abstract: In this paper, we have presented a new multivariate fuzzy time series forecasting method. This method assumes mfactors with one main factor of interest. History of past three years is used for making new forecasts. This new method is applied in forecasting total number of car accidents in Belgium using four secondary factors. We also make comparison of our proposed method with existing methods of fuzzy time series forecasting. Experimentally, it is shown that our proposed method perform better than existing fuzzy time series forecasting methods. Practically, actuaries are interested in analysis of the patterns of causalities in road accidents. Thus using fuzzy time series, actuaries can define fuzzy premium and fuzzy underwriting of car insurance and life insurance for car insurance. National Institute of Statistics, Belgium provides region of risk classification for each road. Thus using this risk classification, we can predict premium rate and underwriting of insurance policy holders.
Abstract: In this study, a fuzzy integrated logical forecasting method (FILF) is extended for multi-variate systems by using a vector autoregressive model. Fuzzy time series forecasting (FTSF) method was recently introduced by Song and Chissom [1]-[2] after that Chen improved the FTSF method. Rather than the existing literature, the proposed model is not only compared with the previous FTS models, but also with the conventional time series methods such as the classical vector autoregressive model. The cluster optimization is based on the C-means clustering method. An empirical study is performed for the prediction of the chartering rates of a group of dry bulk cargo ships. The root mean squared error (RMSE) metric is used for the comparing of results of methods and the proposed method has superiority than both traditional FTS methods and also the classical time series methods.
Abstract: Time series models have been used to make predictions of academic enrollments, weather, road accident, casualties and stock prices, etc. Based on the concepts of quartile regression models, we have developed a simple time variant quantile based fuzzy time series forecasting method. The proposed method bases the forecast using prediction of future trend of the data. In place of actual quantiles of the data at each point, we have converted the statistical concept into fuzzy concept by using fuzzy quantiles using fuzzy membership function ensemble. We have given a fuzzy metric to use the trend forecast and calculate the future value. The proposed model is applied for TAIFEX forecasting. It is shown that proposed method work best as compared to other models when compared with respect to model complexity and forecasting accuracy.
Abstract: In the last 15 years, a number of methods have been proposed for forecasting based on fuzzy time series. Most of the fuzzy time series methods are presented for forecasting of enrollments at the University of Alabama. However, the forecasting accuracy rates of the existing methods are not good enough. In this paper, we compared our proposed new method of fuzzy time series forecasting with existing methods. Our method is based on frequency density based partitioning of the historical enrollment data. The proposed method belongs to the kth order and time-variant methods. The proposed method can get the best forecasting accuracy rate for forecasting enrollments than the existing methods.
Abstract: The paper presents a method for multivariate time
series forecasting using Independent Component Analysis (ICA), as a preprocessing tool. The idea of this approach is to do the forecasting in the space of independent components (sources), and then to transform back the results to the original time series
space. The forecasting can be done separately and with a different
method for each component, depending on its time structure. The
paper gives also a review of the main algorithms for independent component analysis in the case of instantaneous mixture models, using second and high-order statistics. The method has been applied in simulation to an artificial multivariate time series
with five components, generated from three sources and a mixing matrix, randomly generated.
Abstract: ANNARIMA that combines both autoregressive integrated moving average (ARIMA) model and artificial neural network (ANN) model is a valuable tool for modeling and forecasting nonlinear time series, yet the over-fitting problem is more likely to occur in neural network models. This paper provides a hybrid methodology that combines both radial basis function (RBF) neural network and auto regression (AR) model based on binomial smoothing (BS) technique which is efficient in data processing, which is called BSRBFAR. This method is examined by using the data of Canadian Lynx data. Empirical results indicate that the over-fitting problem can be eased using RBF neural network based on binomial smoothing which is called BS-RBF, and the hybrid model–BS-RBFAR can be an effective way to improve forecasting accuracy achieved by BSRBF used separately.
Abstract: Since the pioneering work of Zadeh, fuzzy set theory has been applied to a myriad of areas. Song and Chissom introduced the concept of fuzzy time series and applied some methods to the enrollments of the University of Alabama. In recent years, a number of techniques have been proposed for forecasting based on fuzzy set theory methods. These methods have either used enrollment numbers or differences of enrollments as the universe of discourse. We propose using the year to year percentage change as the universe of discourse. In this communication, the approach of Jilani, Burney, and Ardil is modified by using the year to year percentage change as the universe of discourse. We use enrollment figures for the University of Alabama to illustrate our proposed method. The proposed method results in better forecasting accuracy than existing models.
Abstract: Model Predictive Control has been previously applied
to supply chain problems with promising results; however hitherto
proposed systems possessed no information on future demand. A
forecasting methodology will surely promote the efficiency of
control actions by providing insight on the future. A complete supply
chain management framework that is based on Model Predictive
Control (MPC) and Time Series Forecasting will be presented in this
paper. The proposed framework will be tested on industrial data in
order to assess the efficiency of the method and the impact of
forecast accuracy on overall control performance of the supply chain.
To this end, forecasting methodologies with different characteristics
will be implemented on test data to generate forecasts that will serve
as input to the Model Predictive Control module.
Abstract: Time series forecasting is an important and widely
popular topic in the research of system modeling. This paper
describes how to use the hybrid PSO-RLSE neuro-fuzzy learning
approach to the problem of time series forecasting. The PSO
algorithm is used to update the premise parameters of the
proposed prediction system, and the RLSE is used to update the
consequence parameters. Thanks to the hybrid learning (HL)
approach for the neuro-fuzzy system, the prediction performance
is excellent and the speed of learning convergence is much faster
than other compared approaches. In the experiments, we use the
well-known Mackey-Glass chaos time series. According to the
experimental results, the prediction performance and accuracy in
time series forecasting by the proposed approach is much better
than other compared approaches, as shown in Table IV. Excellent
prediction performance by the proposed approach has been
observed.