An evolving algorithm based on unobservable components neuro-fuzzy model for time series forecasting

Short-term load forecasting method based on fuzzy time series, seasonality and long memory process

Seasonal behaviours are widely encountered in various applications. For instance, requests on web servers are highly influenced by our daily activities. Seasonal forecasting consists in forecasting the whole next season for a given seasonal time series. It may help a service provider to provision correctly the potentially required resources, avoiding critical situations of over- or under provision. In this article, we propose a generic framework to make seasonal time series forecasting. The framework combines machine learning techniques (1) to identify the typical seasons and (2) to forecast the likelihood of having a season type in one season ahead. We study this framework by comparing the mean squared errors of forecasts for various settings and various datasets. The best setting is then compared to state-of-the-art time series forecasting methods. We show that it is competitive with them.

Problem statement: The aim of this research was to study further some latest progress of wavelet transform for time series forecasting, particularly about Neural Networks Multiscale Autoregressive (NN-MAR). Approach: There were three main issues that be considered further in this research. The first was some properties of scale and wavelet coefficients from Maximal Overlap Discrete Wavelet Transform (MODWT) decomposition, particularly at seasonal time series data. The second focused on the development of model building procedures of NN-MAR based on the properties of scale and wavelet coefficients. Then, the third was empirical study about the implementation of the proposed procedure and comparison study about the forecast accuracy of NN-MAR to other forecasting models. Results: The results showed that MODWT at seasonal time series data also has seasonal pattern for scale coefficient, whereas the wavelet coefficients are stationer. The result of model building procedure development yielded a new proposed procedure of NN-MAR model for seasonal time series forecasting. In general, this procedure accommodated input lags of scale and wavelet coefficients and other additional seasonal lags. In addition, the result showed that the proposed procedure works well for determining the best NN-MAR model for seasonal time series forecasting. Conclusion: The comparison study of forecast accuracy showed that the NN-MAR model yields better forecast than MAR and ARIMA models.

Seasonality is a distinctive characteristic which is often observed in many practical time series. Artificial Neural Networks (ANNs) are a class of promising models for efficiently recognizing and forecasting seasonal patterns. In this paper, the Particle Swarm Optimization (PSO) approach is used to enhance the forecasting strengths of feedforward ANN (FANN) as well as Elman ANN (EANN) models for seasonal data. Three widely popular versions of the basic PSO algorithm, viz. Trelea-I, Trelea-II and Clerc-Type1 are considered here. The empirical analysis is conducted on three real-world seasonal time series. Results clearly show that each version of the PSO algorithm achieves notably better forecasting accuracies than the standard Backpropagation (BP) training method for both FANN and EANN models. The neural network forecasting results are also compared with those from the three traditional statistical models, viz. Seasonal Autoregressive Integrated Moving Average (SARIMA), Holt-Winters (HW) and Support Vector Machine (SVM). The comparison demonstrates that both PSO and BP based neural networks outperform SARIMA, HW and SVM models for all three time series datasets. The forecasting performances of ANNs are further improved through combining the outputs from the three PSO based models.

Nonlinear, Non-stationary and Seasonal Time Series Forecasting Using Different Methods Coupled with Data Preprocessing

Triple exponential smoothing is one of the prominent linear models for seasonal time series forecasting. Fuzzy time series forecasting is originated as a new advent for forecasting the data which is imprecise and vague. In this work, we proposed a methodology using both triple exponential smoothing and fuzzy time series. It has the advantage of modelling aspects in linear and non-linear setup. Empirical results with real-world data sets show that the hybrid model is an efficient one based on forecasting accuracy than the component models used individually.KeywordsFuzzy time seriesHybridTriple exponential smoothing

In the paper the algorithm of singular structural analysis and forecasting of a multidimensional series by the MSSA method is used. The program was developed, in which the steps of the method for the selection of singular decomposition components were implemented, the analysis and forecast of real time series was carried out. Singular spectrum analysis (SSA) is increasingly used in the study of time series. Unlike other methods of statistical research of time series, this method is used to study the structure, selection of individual components and forecast of both stationary and non-stationary time series. It does not require an analytical model of the series. In fact, this approach is based on the method of principal components. It is based on the transformation of a series into a matrix and its singular decomposition. After identifying the components of the singular decomposition, their grouping takes place, which leads to the decomposition of the original series into additive components, such as trend, oscillations (periodics), and noise. The SSA method allows you to continue the structure of the time series, thus building a forecast (continuation). An important direction of the development of the SSA method as a method of time series analysis is its generalization for the analysis of multidimensional time series. The method is known as MSSA (Multi-Channel SSA) or E-EOFs (Extended Empirical Orthogonal Functions). In this case, the expected result is the simultaneous decomposition of several series into interpreted components. However, a sufficiently complete theory for MSSA does not exist.

The merits of the modelling philosophy of Box \& Jenkins (1970) are illustrated with a summary of our recent work on seasonal river flow forecasting. Specifically, this work demonstrates that the principle of parsimony, which has been questioned by several authors recently, is helpful in selecting the best model for forecasting seasonal river flow. Our work also demonstrates the importance of model adequacy. An adequate model for seasonal river flow must incorporate seasonal periodic correlation. The usual autoregressive-moving average (ARMA) and seasonal ARMA models are not adequate in this respect for seasonal river flow time series. A new diagnostic check, for detecting periodic correlation in fitted ARMA models is developed in this paper. This diagnostic check is recommended for routine use when fitting seasonal ARMA models. It is shown that this diagnostic check indicates that many seasonal economic time series also exhibit periodic correlation. Since the standard forecasting methods are inadequate on this account, it can be concluded that in many cases, the forecasts produced are sub-optimal. Finally, a limitation of the arbitrary combination of forecasts is also illustrated. Combining forecasts from an adequate parsimonious model with an inadequate model did not improve the forecasts whereas combining the two forecasts of two inadequate models did yield an improvement in forecasting performance. These findings also support the model building philosophy of Box \& Jenkins. The non-intuitive findings of Newbold \& Granger (1974) and Winkler \& Makridakis (1983) that the apparent arbitrary combination of forecasts from similar models will lead to forecasting performance is not supported by our case study with river flow forecasting.

There have been several researches of applying Deep Belief Networks (DBNs) to predict time series data. Most of these works pointed out that DBNs can bring out better prediction accuracy than traditional Artificial Neural Networks. However, one of the main shortcomings of using DBNs in time series prediction concerns with the proper selection of their parameters. In this paper, we investigate the use of Harmony Search algorithm for determining the parameters of DBN in forecasting time series. Experimental results on several synthetic and real world time series datasets revealed that the DBN with parameters selected by Harmony Search performs better than the DBN with parameters selected by Particle Swarm Optimization (PSO) or random method in most of the tested datasets.

The identification of nonlinear systems with artificial neural networks models has been successfully used in many applications. Most processes in industry are characterized by nonlinear and time-varying behavior. In this context, the identification of mathematical models for nonlinear systems is vital in many fields of engineering. The Radial Basis Function Neural Network (RBF-NN) is a powerful approach for nonlinear identification and can be improved using Particle Swarm Optimization (PSO) approaches. This paper presents a multivariable nonlinear system identification using RBF-NN combined with standard PSO and Constriction Factor PSO (CFPSO) approaches in order to determine the RBF-NN parameters. RBF-NN is considered to be a good choice for black-box modeling problems due to its rapid learning capacity and, therefore, has been applied successfully to nonlinear time series modeling and nonlinear identification. On the other hand, PSO was inspired by the choreography of bird flocks and fish schools and can be seen as a distributed behavior algorithm that performs multidimensional search. Furthermore, promising simulation results from performance analysis of the proposed RBF-NN with PSO training approaches are presented and discussed in this paper showing promising results.

Exponomial forecasts of nonstationary time series

In this work an algorithm to adjust parameters using a Bayesian method for cumulative rainfall time series forecasting implemented by an ANN-filter is presented. The criterion of adjustment comprises to generate a posterior probability distribution of time series values from forecasted time series, where the structure is changed by considering a Bayesian inference. These are approximated by the ANN based predictor in which a new input is taken in order for changing the structure and parameters of the filter. The proposed technique is based on the prior distribution assumptions. Predictions are obtained by weighting up all possible models and parameter values according to their posterior distribution. Furthermore, if the time series is smooth or rough, the fitting algorithm can be changed to suit, in function of the long or short term stochastic dependence of the time series, an on-line heuristic law to set the training process, modify the NN topology, change the number of patterns and iterations in addition to the Bayesian inference in accordance with Hurst parameter H taking into account that the series forecasted has the same H as the real time series. The performance of the approach is tested over a time series obtained from samples of the Mackey-Glass delay differential equations and cumulative rainfall time series from some geographical points of Cordoba, Argentina.

Non-stationary time series plays a prominent role in the analysis of performance time series of many real-world systems. Recently, fuzzy time series models have been extended to forecast non-stationary time series. Over different phases of time, performance time series may show drastic changes. Therefore, a non-stationary time series is partitioned according to different phases of time. These phases may be taken as weeks, months, or years. Over different phases of time, the universe of discourse, knowledge base, and rule base may vary. A common constraint in the modelling of time-variant fuzzy time series is their incapability to address the phase change of time in the model and accordingly incorporate the necessary changes in the universe of discourse, knowledge base, and rule base over different time phases. To address this issue, a Phase-cum-Time Variant Fuzzy Time Series Model (PTVFTS) is presented in this paper. The PTVFTS model is developed so that it will address the problem of phase change as well as the time variations within each phase simultaneously. The concept of the model rebuilding process is applied to handle changes over each phase and the modified parameter adaptation technique is used for the time variations within each phase. The developed model is applied to the daily closing price time series for the years 2017, 2018, 2019, 2020, 2021, and 2022 separately of stock market indices, NASDAQ, S&P 500, Dow Jones, and TAIEX. The comparison of the developed model is made with the time-variant fuzzy time series model known as the non-stationary fuzzy time series model (NSFTS), Dynamic Evolving Neural-Fuzzy Inference System (DENFIS) model, Long Short-Term Memory (LSTM) model, and the classical Auto-Regressive Integrated Moving Average (ARIMA) model. The efficiency of the developed PTVFTS model is tested using forecasting metrics and statistical tests. The comparison shows that the developed model is more efficient in forecasting phase-cum-time variant non-stationary time series.

Fu zzy time series forecasting methods have been widely studied in recent years. This is because fuzzy time series forecasting methods are co mpatib le with flexib le calculat ion techniques and they do not require constraints that exist in conventional time series approaches. Most of the real life time series exhibit periodical changes arising fro m seasonality. These variations are called seasonal changes. Although, conventional time series approaches for the analysis of time series which have seasonal effect are abundant in literature, the number of fuzzy t ime series approaches is limited. In almost all of these studies, membership values are ignored in the analysis process. This affects forecasting performance of the approach negatively due to the loss of information as well as posing a situation that is incompatible with the basic features of fuzzy set theory. In this study, for the first time in literature, a new seasonal fuzzy time series approach which considers membership values in both identificat ion of fuzzy relat ions and defuzzification steps was proposed. In the proposed method, we used fuzzy C-means clustering method in fuzzification step and artificial neural networks (ANN) in identification of fu zzy relation and defuzzification steps which consider membership values. The proposed method was applied to various seasonal fuzzy time series and obtained results were compared with some conventional and fuzzy time series approaches. In consequence of this evaluation, it was determined that forecasting performance of the proposed method is satisfactory.

В работе предложены математические модели прогнозирования нестационарных временных рядов с учётом внешних факторов и методы их структурной идентификации, основанные на совместном использовании многомерного варианта метода «Гусеница»-SSA и моделей SARIMAX, расширенных для прогнозирования временных рядов с несколькими сезонными составляющими и с учётом нескольких экзогенных переменных. Экспериментальные результаты показывают высокую эффективность предложенных моделей прогнозирования при выборе подходящих структурных параметров в сравнении с моделями SARIMAX.