Classical Time Series Research Articles

Probabilistic Forecasting of Wind and Solar Farm Output

Accurately forecasting the output of grid connected wind and solar systems is critical to increasing the overall penetration of renewables on the electrical network. This includes not only forecasting the expected level, but also putting error bounds on the forecast. The National Electricity Market (NEM) in Australia operates on a five minute basis. We used statistical forecasting tools to generate forecasts with prediction intervals, trialing them on one wind and one solar farm. In classical time series forecasting, construction of prediction intervals is rudimentary if the error variance is constant—Termed homoscedastic. However, if the variance changes—Either conditionally as with wind farms, or systematically because of diurnal effects as with solar farms—The task is much more complicated. The tools were trained on segments of historical data and then tested on data not used in the training. Results from the testing set showed good performance using metrics, including Coverage and Interval Score. The methods used can be adapted to various time scales for short term forecasting.

Forecasting under Long Memory

Abstract Motivated by the mixed evidence in the literature on forecasting long memory processes, we show that methods based on fractional integration are superior to alternatives not accounting for long memory by simulations and applications to classical long memory time series from macroeconomics and finance. Furthermore, we analyze the optimal implementation of these methods, among others comparing parametric and local and global semiparametric estimators of the long memory parameter, providing asymptotic theory on different mean estimators and assessing the use of a fixed long memory parameter to overcome the inherent difficulties of its estimation.

Deposit and Withdrawal Dynamics: A Data-Based Mutually-Exciting Stochastic Model

This paper proposes a novel self- and mutual-exciting stochastic model to capture two essential features underlying a general type of discrete-time event data motivated by the practice: the dependence on the past event arrivals and associated sizes (i.e., self-exciting) and the behavioral interdependence between multiple activities (i.e., mutual-exciting). Despite the technical challenges in capturing these two features under a general discrete-time framework, we are able to fully characterize the probability distribution for the proposed model (i.e., the closed-form characteristic functions), theoretically quantify customer performance measures, and establish efficient maximum likelihood estimation. To illustrate the applicability of our model and validate its performance, we calibrate it with a customer deposit and withdrawal data set from one leading online money market fund. We analytically quantify the churn probability and expected activity level of customers as an illustration of performance measures and then compare our model with classic time-series and machine-learning models and show that our model can achieve high prediction accuracy. The theoretical tractability and predictive accuracy of the proposed framework enable us to build optimization models to improve firm performance, and we illustrate one application through a personalized interest-rate optimization problem. On a broader note, our model framework is generally applicable to characterize any discrete-time event data with self and mutual excitation in nature and to inform optimal policies for decision-makers.

EXCHANGE RATE FORECASTING USING FUZZY TIME SERIES-MARKOV CHAIN

Exchange rate forecasting plays an important role in financial management. However, it is a complex process with high nonlinearity and data irregularity. Moreover, the forecasting of exchange rate is highly involved with imprecise and uncertain data. Analysis of forecasting models which corresponds to the exchange rate has always experienced fluctuations. Therefore, exchange rate forecasting becomes a challenging task in finance. Several studies have shown that stand-alone forecasting models such as time series, fuzzy time series, and Markov chain have their own drawbacks and are not successful enough in forecasting accurately. In this study, we propose a hybrid model of fuzzy time series-Markov chain to forecast the future exchange rate. Fuzzy time series-Markov chain is a combination of the classic fuzzy time series model with Markov chain model used to analyse a set of time series data. The main motivation for this study is to improve the accuracy in exchange rate forecasting. The selected currencies are Malaysian Ringgit (MYR) and Singapore Dollar (SGD). The proposed model was then evaluated by the Mean Absolute Percentage Error (MAPE) performance metric to test the robustness of the model. Lastly, a comparison between the proposed model and fuzzy time series model was conducted with respect to the MAPE. The results showed that the MAPE value for fuzzy time series-Markov chain was 0.9895% which fell under the criterion of highly accurate forecasting. Meanwhile, the MAPE value for fuzzy time series was 3.4306%. Thus, the forecasting performance of the proposed model was better than the fuzzy time series model. This study reveals the potential benefits of the proposed model as a highly accurate forecasting model.

A Transfer Entropy Based Approach for Fault Isolation in Industrial Robots

Abstract In this paper, we cast the problem of fault isolation in industrial robots as that of causal analysis within coupled dynamical processes and evaluate the related efficacy of the information-theoretic approach of transfer entropy. To create a realistic and exhaustive dataset, we simulate wear-induced failure by increasing friction coefficient on select axes within an in-house robotic simulation tool that incorporates an elastic gearbox model. The source axis of failure is identified as one which has the highest net transfer entropy across all pairs of axes. In an exhaustive simulation study, we vary the friction successively in each axis across three common industrial tasks: pick and place, spot welding, and arc welding. Our results show that transfer entropy-based approach is able to detect the axis of failure more than 80% of the time when the friction coefficient is 5% above the nominal value and always when friction coefficient is 10% above the nominal value. The transfer entropy approach is more than twice as accurate as cross-correlation, a classical time series analysis used to identify directional dependence among processes.

A parallel-series hybridization of seasonal intelligent based statistical model for demand forecasting

PurposeThe purpose of this paper, because of the complexity of demand time series and the need to construct a more accurate hybrid model that can model all relationships in data, is to propose a parallel-series hybridization of seasonal neural networks and statistical models for demand time series forecasting.Design/methodology/approachThe main idea of proposed model is centered around combining parallel and series hybrid methodologies to use the benefit of unique advantages of both hybrid strategies as well as intelligent and classic seasonal time series models simultaneously for achieving results that are more accurate for the first time. In the proposed model, in contrast of traditional parallel and series hybrid strategies, it can be generally shown that the performance of the proposed model will not be worse than components.FindingsEmpirical results of forecasting two well-known seasonal time series data sets, including the total production value of the Taiwan machinery industry and the sales volume of soft drinks, indicate that the proposed model can effectively improve the forecasting accuracy achieved by either of their components used in isolation. In addition, the proposed model can achieve more accurate results than parallel and series hybrid model with same components. Therefore, the proposed model can be used as an appropriate alternative model for seasonal time series forecasting, especially when higher forecasting accuracy is needed.Originality/valueTo the best of the authors’ knowledge, the proposed model, for first time and in contrast of traditional parallel and series hybrid strategies, is developed.

A TensorFlow Approach to Data Analysis for Time Series Forecasting in the Energy-Efficiency Realm

Thanks to advances in smart metering devices (SM), the electricity sector is undergoing a series of changes, among which it is worth highlighting the ability to control the response to all events that occur in the electricity grid with the intention of making it more smart. Predicting electricity consumption data is a key factor for the energy sector in order to create a completely intelligent electricity grid that optimizes consumption and forecasts future energy needs. However, it is currently not enough to give a prediction of energy consumption (EC), but it is also necessary to give the prediction as fast as possible so that the grid can operate in the shortest possible time. An approach for developing EC prediction systems is introduced here by the use of artificial neural networks (ANN). Differently from other research studies on the subject, a divide-and-conquer strategy is used so that the target system’s execution switches from one to another specialized small models that forecast the EC of a building within the time range of one hour. By simultaneously processing a large amount of data and models, a consequence of implementing them in parallel with TensorFlow on GPUs, the training procedure proposed here increases the performance of the classic time series prediction methods, which are based on ANN. Leveraging the latest generation of ANN techniques and new GPU-based architectures, correct EC predictions can be obtained and, as the experimentation carried out in this work shows, such predictions can be obtained quickly. The obtained results in this study show a promising way for speeding up big data processing of building’s monitoring data to achieve energy efficiency.

Forecasting Currency in East Java: Classical Time Series vs. Machine Learning

Most research about the inflow and outflow currency in Indonesia showed that these data contained both linear and nonlinear patterns with calendar variation effect. The goal of this research is to propose a hybrid model by combining ARIMAX and Deep Neural Network (DNN), known as hybrid ARIMAX-DNN, for improving the forecast accuracy in the currency prediction in East Java, Indonesia. ARIMAX is class of classical time series models that could accurately handle linear pattern and calendar variation effect. Whereas, DNN is known as a machine learning method that powerful to tackle a nonlinear pattern. Data about 32 denominations of inflow and outflow currency in East Java are used as case studies. The best model was selected based on the smallest value of RMSE and sMAPE at the testing dataset. The results showed that the hybrid ARIMAX-DNN model improved the forecast accuracy and outperformed the individual models, both ARIMAX and DNN, at 26 denominations of inflow and outflow currency. Hence, it can be concluded that hybrid classical time series and machine learning methods tend to yield more accurate forecasts than individual models, both classical time series and machine learning methods.

The Deep Learning LSTM and MTD Models Best Predict Acute Respiratory Infection among Under-Five-Year Old Children in Somaliland

The most effective techniques for predicting time series patterns include machine learning and classical time series methods. The aim of this study is to search for the best artificial intelligence and classical forecasting techniques that can predict the spread of acute respiratory infection (ARI) and pneumonia among under-five-year old children in Somaliland. The techniques used in the study include seasonal autoregressive integrated moving averages (SARIMA), mixture transitions distribution (MTD), and long short term memory (LSTM) deep learning. The data used in the study were monthly observations collected from five regions in Somaliland from 2011–2014. Prediction results from the three best competing models are compared by using root mean square error (RMSE) and absolute mean deviation (MAD) accuracy measures. Results have shown that the deep learning LSTM and MTD models slightly outperformed the classical SARIMA model in predicting ARI values.

English

One of the most research discussed topics is the prediction or forecasting of the COVID-19 data using the classical time series (such as exponential smoothing) and machine learning methods. In fact, the classical time series method often produces quite large error rates. In this study, the researchers try to use nonparametric modeling with the kernel method to get better results with the smallest error rate. Furthermore, the results of the kernel method are compared with the results of the classical time series method. As a comparison tool, the researchers use MAPE by paying attention to the smallest MAPE value. The data used in this study are the COVID-19 data in Indonesia in which its variable is the total of deaths per day. After comparing the classical time series method with the kernel method, the obtained better results are the results from the kernel method. In this study, the researchers use five kernel functions, namely the Gaussian, Epanechnikov, Triangular, Biweight, and Triweight. Then, these five kernel functions are compared to find the best function. After the comparison process is done, the triweight kernel function was determined as the best function with the smallest error rate with a MAPE value of 0,9%. S Keywords — Covid-19, Kernel Method, Mean Absolute Percentage Error, Time Series, Triweight Kernel Function.

Time-aware cloud manufacturing service selection using unknown QoS prediction and uncertain user preferences

Selecting the most efficient from several functionally equivalent services remains an ongoing challenge. Most manufacturing service selection methods regard static quality of service (QoS) as a major competitiveness factor. However, adaptations are difficult to achieve when variable network environment has significant impact on QoS performance stabilization in complex task processes. Therefore, dynamic temporal QoS values rather than fixed values are gaining ground for service evaluation. User preferences play an important role when service demanders select personalized services, and this aspect has been poorly investigated for temporal QoS-aware cloud manufacturing (CMfg) service selection methods. Furthermore, it is impractical to acquire all temporal QoS values, which affects evaluation validity. Therefore, this paper proposes a time-aware CMfg service selection approach to address these issues. The proposed approach first develops an unknown-QoS prediction model by utilizing similarity features from temporal QoS values. The model considers QoS attributes and service candidates integrally, helping to predict multidimensional QoS values accurately and easily. Overall QoS is then evaluated using a proposed temporal QoS measuring algorithm which can self-adapt to user preferences. Specifically, we employ the temporal QoS conflict feature to overcome one-sided user preferences, which has been largely overlooked previously. Experimental results confirmed that the proposed approach outperformed classical time series prediction methods, and can also find better service by reducing user preference misjudgments.

Singular spectrum analysis for the investigation of structural vibrations

This study aims to demonstrate some capabilities of singular spectrum analysis (SSA) for the purposes of analysis of vibrating/dynamic structures. SSA is a powerful method for time series analysis, which encompasses elements of classical time series analysis, multivariate statistics, principal component analysis, dynamical systems and signal processing. This paper introduces the application of SSA for vibration response time series and investigates the decomposition of the structural vibrations into harmonics, trend and noise components and their reconstruction.After an introduction to SSA and its application for structural vibration signals, the paper investigates a simple model of a dynamic system and attempts to analyse it using classical methods of structural dynamics and SSA. The results and the capabilities of the two methods are then compared and discussed. Furthermore, the paper considers an experimental case of a layered GFRP composite plate which is subjected to free vibration and the recorded signals are analysed using conventional structural dynamics methods and SSA. The results demonstrate that SSA can be used as a powerful tool for analysis of the vibratory behaviour especially for structures with more complex and/or nonlinear dynamic behaviour. It is capable of extracting the vibratory/oscillation patterns even in the case of nonlinear oscillations.

Multi-scale information fusion model for feature extraction of converter transformer vibration signal

The vibration signal feature extraction model of the converter transformer has the problem of low accuracy. The main reason is that the time-related information is easily lost during training. In recent years, convolutional neural networks (CNN) have been proven to break the limits of traditional classification models. CNN has achieved remarkable results in large-scale image recognition tasks. To solve the problem of difficulty in model building and loss of time-related information, this paper proposes a multi-scale fusion feature extraction model for converter transformer vibration signal. The model includes an image generation module and a multi-scale information fusion module. The image generation module converts vibration signals into time-domain, frequency-domain, and energy feature images by calculating the Markov Transform Field and Continuous Wavelet Transform. In the multi-scale information fusion module, the attention module is introduced into the multi-parallel convolutional neural network to fuse the feature images. The proposed method can make full use of the advantages of deep learning through generating images that can represent time series from multi-scale. To evaluate the effectiveness of the model in this paper, we establish a data set based on the measured vibration signals and tested by classification experiments. The result shows the accuracy of the model state recognition in this paper is 96.15%, which is better than classic time series processing networks such as LSTM and 1D-CNN.

Long-Term Data Traffic Forecasting for Network Dimensioning in LTE with Short Time Series

Network dimensioning is a critical task in current mobile networks, as any failure in this process leads to degraded user experience or unnecessary upgrades of network resources. For this purpose, radio planning tools often predict monthly busy-hour data traffic to detect capacity bottlenecks in advance. Supervised Learning (SL) arises as a promising solution to improve predictions obtained with legacy approaches. Previous works have shown that deep learning outperforms classical time series analysis when predicting data traffic in cellular networks in the short term (seconds/minutes) and medium term (hours/days) from long historical data series. However, long-term forecasting (several months horizon) performed in radio planning tools relies on short and noisy time series, thus requiring a separate analysis. In this work, we present the first study comparing SL and time series analysis approaches to predict monthly busy-hour data traffic on a cell basis in a live LTE network. To this end, an extensive dataset is collected, comprising data traffic per cell for a whole country during 30 months. The considered methods include Random Forest, different Neural Networks, Support Vector Regression, Seasonal Auto Regressive Integrated Moving Average and Additive Holt–Winters. Results show that SL models outperform time series approaches, while reducing data storage capacity requirements. More importantly, unlike in short-term and medium-term traffic forecasting, non-deep SL approaches are competitive with deep learning while being more computationally efficient.

Spatio-Temporal Traffic Flow Prediction in Madrid: An Application of Residual Convolutional Neural Networks

Due to the need to predict traffic congestion during the morning or evening rush hours in large cities, a model that is capable of predicting traffic flow in the short term is needed. This model would enable transport authorities to better manage the situation during peak hours and would allow users to choose the best routes for reaching their destinations. The aim of this study was to perform a short-term prediction of traffic flow in Madrid, using different types of neural network architectures with a focus on convolutional residual neural networks, and it compared them with a classical time series analysis. The proposed convolutional residual neural network is superior in all of the metrics studied, and the predictions are adapted to various situations, such as holidays or possible sensor failures.

Multiscale attention-based LSTM for ship motion prediction

Ship motion prediction is applied to the shipboard stabilized platform to keep the equipment on the platform stable all the time, which is of great practical significance to the safety and efficiency of shipboard equipment operation. Long Short-term Memory (LSTM) Network is a classic time series prediction method that has made remarkable achievements in this field. However, the dynamic frequency range of single LSTM in ship motion prediction is insufficient to meet the stabilized platform with higher precision requirements. To improve the performance of LSTM in ship motion prediction, this paper presents a novel method named as multiscale attention-based LSTM. At first, wavelet transform is employed to decompose ship motion signals into several frequency scales, which makes LSTM to capture the inherent law of ship motion from each frequency scale. And then the weights of different scales are obtained by attention mechanism, which promote the sensitivity of the whole system by paying more attention to significant information and suppress the interference of noise signals. Both of the steps form a multiscale attention mechanism, which promote the adaptability and improve the performance of the LSTM. In addition, to avoid being trapped in local optimization, the two-stage training mechanism is designed for model training based on the model structure. Ship motion data are used to evaluate the feasibility and effectiveness. The experiments show that the proposed method achieves better performance compared with other popular methods.

Modelling and applications for non-stationary time series in the presence of trend and period

In this paper, we study the estimation of the non-stationary time series in the presence of trend and period. We write the classical decomposition time series model as a partial linear model with an unknown parameter. Then we approximate the trend term with B-spline and obtain the estimators of the period, periodic sequence and trend. The asymptotic behaviors of the estimators are given in our paper, including the consistency of period estimation and the asymptotic behavior of the estimated periodic sequence and trend. We illustrate the superiority of our method through simulations and the applications to two real data examples show the utility of our method.

MDTP

Traffic prediction has drawn increasing attention for its ubiquitous real-life applications in traffic management, urban computing, public safety, and so on. Recently, the availability of massive trajectory data and the success of deep learning motivate a plethora of deep traffic prediction studies. However, the existing neural-network-based approaches tend to ignore the correlations between multiple types of moving objects located in the same spatio-temporal traffic area, which is suboptimal for traffic prediction analytics. In this paper, we propose a multi-source deep traffic prediction framework over spatio-temporal trajectory data, termed as MDTP. The framework includes two phases: spatio-temporal feature modeling and multi-source bridging. We present an enhanced graph convolutional network (GCN) model combined with long short-term memory network (LSTM) to capture the spatial dependencies and temporal dynamics of traffic in the feature modeling phase. In the multi-source bridging phase, we propose two methods, Sum and Concat, to connect the learned features from different trajectory data sources. Extensive experiments on two real-life datasets show that MDTP i) has superior efficiency, compared with classical time-series methods, machine learning methods, and state-of-the-art neural-network-based approaches; ii) offers a significant performance improvement over the single-source traffic prediction approach; and iii) performs traffic predictions in seconds even on tens of millions of trajectory data. we develop MDTP + , a user-friendly interactive system to demonstrate traffic prediction analysis.

Fuzzy Time Series and Artificial Neural Network: Forecasting Exportation of Natural Rubber in Malaysia

Natural rubber is one of the most important crops in Malaysia alongside palm oil, cocoa, paddy, and pineapple. Being a tropical country, Malaysia is one of the top five exporters and producers of rubber in the world. The purpose of this study is to find the forecasted value of the actual data of the number of exportations of natural rubber by using Fuzzy Time Series and Artificial Neural Network. This study is also conducted to determine the best model by making comparison between Fuzzy Time Series and Artificial Neural Network. Fuzzy Time Series has allowed to overcome a downside where the classical time series method cannot deal with forecasting problem in which values of time series are linguistic terms represented by fuzzy sets. Artificial Neural Network was introduced as one of the systematic tools of modelling which has been forecasting for about 20 years ago. The error measure that was used in this study to make comparisons were Mean Square Error, Root Mean Square Error and Mean Absolute Percentage Error. The results of this study showed that the fuzzy time series method has the smallest error value compared to artificial neural network which means it was more accurate compared to artificial neural network in forecasting exportation of natural rubber in Malaysia.

Modeling of water consumption in Saudi Arabia using classical and modern time series methods

Overpopulation, industrialization, urbanization, and the spreading out of irrigated agricultural lands are the driving forces to increase the demand of water in the Kingdom of Saudi Arabia. Improvement, production, and management of various commodities are one of the objectives of vision 2030 of the Kingdom. In this paper, we seek to find the paramount model for water consumption in Saudi Arabia using classical and modern tools of time series analysis. From the classical time series analysis, we employed multiplicative seasonal autoregressive integrated moving average (SARIMA) model, while functional autoregressive (FAR) model was adopted under modern time series analysis. A data of water consumption from January 2001 to December 2016 were analyzed, and the results were compared using these two methods. This study is the first ever application of FAR method in Saudi Arabia using water consumption data. Statistical analysis exhibits that the best model for water consumption data under classical approach is SARIMA (1,0,2)(2,1,1) [12] while FAR (1) under modern paradigm. A comparison between these methods reveals that FAR modeling is better than SARIMA in training as well as in forecasting of water consumption data in the Kingdom. This kind of modeling is beneficial for future policy implications.