Short-term Time Series Forecasting Research Articles

Exploring Deep Learning Approaches for Short-Term Passenger Demand Prediction

An accurate short-term passenger demand forecast makes a contribution to the coordination of traffic supply and demand. Forecasting the short-term passenger demand for the on-demand transportation service platform is of utmost significance since it might incentivize empty cars to relocate from over-supply regions to over-demand regions. Yet, because spatial, temporal, and exogenous dependencies need to be evaluated concurrently, short-term passenger demand forecasting may be rather difficult. This article aims to investigate several methods that can be utilized to forecast short-term traffic demand, with a primary emphasis on deep learning approaches. We examine varying degrees of temporal aggregation and how these levels affect various architectural configurations. In addition, by analyzing 22 models representing 5 distinct architectural configurations, we illustrate the influence of varying layer configurations within each architecture. The findings indicate that the long-term short memory (LSTM) structures perform the best for short-term time series forecasting, but more complex architectures do not significantly enhance the outcomes. Moreover, considering the spatiotemporal aspects results in an improvement in the prediction of more than fifty percent. In addition, we investigate the vectorization of time, also known as Time2Vec, as a way of embedding to make it possible for a selected algorithm to recognize periodic characteristics in time series, and we show that the outcome is improved by fifteen percent.

Detection of Anomalies in the Operation of a Road Lighting System Based on Data from Smart Electricity Meters

Smart meters in road lighting systems create new opportunities for automatic diagnostics of undesirable phenomena such as lamp failures, schedule deviations, or energy theft from the power grid. Such a solution fits into the smart cities concept, where an adaptive lighting system creates new challenges with respect to the monitoring function. This article presents research results indicating the practical feasibility of real-time detection of anomalies in a road lighting system based on analysis of data from smart energy meters. Short-term time series forecasting was used first. In addition, two machine learning methods were used: one based on an autoregressive integrating moving average periodic model (SARIMA) and the other based on a recurrent network (RNN) using long short-term memory (LSTM). The algorithms were tested on real data from an extensive lighting system installation. Both approaches enable the creation of self-learning, real-time anomaly detection algorithms. Therefore, it is possible to implement them on edge computing layer devices. A comparison of the algorithms indicated the advantage of the method based on the SARIMA model.

Multilinear-Trend Fuzzy Information Granule-Based Short-Term Forecasting for Time Series

Although the idea of information granulation has been shown to be a research craze in short-term time series forecasting, it is still urgent to develop a granular framework so that information granulation can characterize the trend distribution of data to the significant extent under a common concept of time. This article puts forward a novel granulation algorithm involving two-stage partitioning scheme so that information granule established there exhibits well-articulated semantics at the time level, while at the same time, it gives full consideration to the varying patterns of data. On this basis, a new association rule based on this type of information granules is presented. Unlike most fuzzy association rules, the proposed association rules can extract and derive the correlations between two collections of trend features corresponding to the past and future, which is in accord with human's reasoning. Keeping in mind that the prediction process should center on essential rules while freeing from the interference of irrelevant rules, which contributes to a reliable prediction result, thus, a rule selection algorithm is involved so as one makes sure the accuracy and interpretability of the forecasting results. The design of short-term forecasting model based on fuzzy inference system is implemented, where the concept of granulation eliminates the commonly used alternative, i.e., the recursive iterations of one-step prediction. Experimental results have verified the effectiveness of the proposed model.

High granular and short term time series forecasting of $$\hbox {PM}_{2.5}$$ air pollutant - a comparative review

Forecasting time series has acquired immense research importance and has vast applications in the area of air pollution monitoring. This work attempts to investigate the abilities of various existing techniques when applied for short term, high granular time series forecasting of PM2.5. More specifically, a comparative study has been provided, taking into account both popularly used models and lesser-used models in this area. The study has been carried out considering ten well defined models that are ARIMA (auto-regressive integrated moving average), SARIMA (seasonal ARIMA), SES (single exponential smoothing), DES (double exponential smoothing), TES (triple exponential smoothing), ANN (artificial neural network), DT (decision tree), kNN (k-nearest neighbor), LSTM (long short-term memory) and MCFO (markov chain first order). A framework has been built that categories the models, implements them under identical execution environment and forecasts succeeding values. Implementation has been carried out over five data sets of real-world air pollution time series, that are collected from five differently located government setup monitoring stations over a period of 1 year (July 2018-June 2019). Rigorous statistical analysis has been performed that yields an insight to the nature and variability of these time series data. Forecasting has been carried out on short term basis, focusing on high granularity whereas, three different lengths of forecast horizon (1 day, 1 week, and 1 month) have been tested. Eventually, the models have been compared in terms of their associated performance measuring units namely, RMSE (root mean of squared error), MAE (mean absolute error) and MAPE (mean absolute percentage error). The comparative results verified with multiple datasets show that all the models posses less error for a shorter forecast horizon, where LSTM providing the best performance. Superiority of machine learning and deep learning models are found in case of longer length of forecast horizon with kNN achieving best accuracy whereas, significant performance degradation of ARIMA is found for longer forecast horizon. Moreover, TES, DT, kNN, LSTM, MCFO are found to be well adopted in relation with shape and variability of the data. Note that the performance on various length of high granular forecast horizon have been studied over multiple datasets that give an added value to this work.

Predicting the Future is Like Completing a Painting: Towards a Novel Method for Time-Series Forecasting

This article is an introductory work towards a larger research framework relative to Scientific Prediction. It is a mixed between science and philosophy of science, therefore we can talk about Experimental Philosophy of Science. As a first result, we introduce a new forecasting method based on image completion, named Forecasting Method by Image Inpainting (FM2I). In fact, time series forecasting is transformed into fully images- and signal-based processing procedures. After transforming a time series data into its corresponding image, the problem of data forecasting becomes essentially a problem of image inpainting, i.e., completing missing data in the image. Extensive experimental evaluation was conducted using the shortest series of the dataset proposed by the well-known M3-competition. Results show that FM2I, despite still being in its infancy, represents an efficient and robust tool for short-term time series forecasting. It has achieved prominent results in terms of accuracy and outperforms the best M3 methods. We have also investigated the effectiveness of the FM2I against the Smyl method, the winner of the M4 competition. Using the same category of shortest series, results show a close accuracy compared to the Smyl method. The FM2I is also able to generate ensemble data forecasts, which contain the best and more accurate forecast compared to existing and considered methods.

Long- and short-term time series forecasting of air quality by a multi-scale framework.

Air quality forecasting for Hong Kong is a challenge. Even taking the advantages of auto-regressive integrated moving average and some state-of-the-art numerical models, a recently-developed hybrid method for one-day (two- and three-day) ahead forecasting performs similarly to (slightly better than) a simple persistence forecasting. Long-term forecasting also remains an important issue, especially for policy decision for better control of air pollution and for evaluation of the long-term impacts on public health. Given the well-recognized negative effects of PM2.5, NO2 and O3 on public health, we study their time series under the multi-scale framework with empirical mode decomposition and nonstationary oscillation resampling to explore the possibility of long-term forecasting and to improve short-term forecasts in Hong Kong. Applied to a dataset from January 2016 to December 2018, the long-term forecasting (with lead time about 100 days) of the multi-scale framework has the root-mean-square error (RMSE) comparable with that of the short-term (with lead time of one or two days) forecasting by the persistence method, while its improvement for short-term forecasting (with lead time of one, two or three days) is quite substantial over the persistence forecasting, with RMSEs reduced by respectively 44%-47%, 30%-45%, and 40%-60% for PM2.5, NO2, and O3. Compared to the hybrid method, it turns out that, for short-term forecasting for the same data, the multi-scale framework can reduce RMSE by about 25% (respectively 30%) for PM2.5 (respectively NO2 and O3). In addition, we find no significant difference in the forecasting performance of the multi-scale framework among different types of stations. The multi-scale framework is feasible for time series forecasting and applicable to other pollutants in other cities.

COVID-19 with Uncertain Phases: Estimation Issues with An Illustration for Argentina

We use an approach to assess COVID-19 performance that starts from what we consider is the most likely set of hypotheses about the uncertain evolution of the pandemic, that envisage a sequence of different cycles with unknown duration and magnitude over 18-24 months. This pattern implies a research strategy where short-term time series forecasting of the evolution of observed cases and deaths play a central role in both detecting transitions from phase to phase of infections and the estimation of necessarily changing structural parameters and indicators of a SIRD model. We illustrate our approach with Buenos Aires City performance, which represents a significant share of the Argentine case with an early introduction of a lockdown followed by a second wave latter on. This approach can be extended to include measures of the intensity and compliance of lockdowns, as well as the heterogeneity across areas. We find that mobility (as a proxy for the effectiveness of the lockdown) has an impact on observed cases in Buenos Aires City with a lag of 8 days and deaths relate with new cases registered 16 to 19 days before. Mobility has a clear impact on the growth rate of cases and by extension deaths. Our approach and results have implications for policy dialogue issues.

A New Approach for Short-term Time Series Forecasting

Short-term series forecasting is one of the essential issues in a variety of tasks, such as traffic flow prediction, stock market tendency analysis, etc. Most current methods based on stable or abundant historical data. In this paper, we proposed a novel model called LA-NN. It takes advantages of both long short-term memory (LSTM) network and autoregressive integrated moving average (ARIMA) by a relation integration of them. So as to deal with the situation of insufficient historical data and sudden abnormal changes in data. A comparison with other representative forecast models validates that the proposed LA-NN network can achieve a better performance.

AI-based Short-Term Electric Time Series Forecasting

In current scenario of various electrical profiles like load profile, energy met profile, peak demand, etc. are very complex and therefore affected proper power system planning. Electrical forecasting is an important part in proper power system planning. Classical models, i.e., time series models and other conventional models are restricted for linear and stationary electrical profiles. Consequently, these models are not suitable for accurate electrical forecasting. In this paper, artificial neural network (ANN) based forecasting models are proposed to forecast hydro generation, energy met and peak demand. Auto-regressive (AR), moving average (MA), Auto-regressive Moving average (ARMA) and auto-regressive integrated moving average (ARIMA) are also developed to show the effectiveness of ANN based models over time series models. Additionally, best selection of hidden neurons in ANN is also shown here.

Sparse short-term time series forecasting models via minimum model complexity

Time series forecasting is of fundamental importance in big data analysis. The prediction of noisy, non-stationary, and chaotic time series demands good generalization from small amounts of data. Vapnik showed that the total risk is dependent on both, empirical error as well as model complexity, where the latter may be measured in terms of the Vapnik–Chervonenkis (VC) dimension. In other words, good generalization requires minimizing model complexity. The recently proposed Minimal Complexity Machine (MCM) has been shown to minimize a tight bound on the VC dimension, and has further been extended to Minimal Complexity Machine Regression (MCMR). In this paper, we present an original approach based on the MCM regressor, which builds sparse and accurate models for short-term time series forecasting. Results on a number of datasets establish that the proposed approach is superior to a number of state-of-the-art methods, and yields sparse models. These sparse models are able to extract only the most important information present in the data sets, thereby achieving high accuracy. Sparsity in time series forecasting models is also important in reducing the evaluation time. This assumes importance when the models need to be evaluated in real time, such as when they are used as part of trading flows.

SHORT-TERM SOLAR RADIATION FORECASTING BY USING AN ITERATIVE COMBINATION OF WAVELET ARTIFICIAL NEURAL NETWORKS

The information provided by accurate forecasts of solar energy time series are considered essential for performing an appropriate prediction of the electrical power that will be available in an electric system, as pointed out in Zhou et al. (2011). However, since the underlying data are highly non-stationary, it follows that to produce their accurate predictions is a very difficult assignment. In order to accomplish it, this paper proposes an iterative Combination of Wavelet Artificial Neural Networks (CWANN) which is aimed to produce short-term solar radiation time series forecasting. Basically, the CWANN method can be split into three stages: at first one, a decomposition of level p, defined in terms of a wavelet basis, of a given solar radiation time series is performed, generating r+1 Wavelet Components (WC); at second one, these r+1 WCs are individually modeled by the k different ANNs, where k>5, and the 5 best forecasts of each WC are combined by means of another ANN, producing the combined forecasts of WC; and, at third one, the combined forecasts WC are simply added, generating the forecasts of the underlying solar radiation data. An iterative algorithm is proposed for iteratively searching for the optimal values for the CWANN parameters, as we will see. In order to evaluate it, ten real solar radiation time series of Brazilian system were modeled here. In all statistical results, the CWANN method has achieved remarkable greater forecasting performances when compared with a traditional ANN (described in Section 2.1).

Markov Chain Modelling for Short-Term NDVI Time Series Forecasting

In this paper, the NDVI time series forecasting model has been developed based on the use of discrete time, continuous state Markov chain of suitable order. The normalised difference vegetation index (NDVI) is an indicator that describes the amount of chlorophyll (the green mass) and shows the relative density and health of vegetation; therefore, it is an important variable for vegetation forecasting. A Markov chain is a stochastic process that consists of a state space. This stochastic process undergoes transitions from one state to another in the state space with some probabilities. A Markov chain forecast model is flexible in accommodating various forecast assumptions and structures. The present paper discusses the considerations and techniques in building a Markov chain forecast model at each step. Continuous state Markov chain model is analytically described. Finally, the application of the proposed Markov chain model is illustrated with reference to a set of NDVI time series data.

Predicting short-term bus passenger demand using a pattern hybrid approach

This paper proposes an Interactive Multiple Model-based Pattern Hybrid (IMMPH) approach to predict short-term passenger demand. The approach maximizes the effective information content by assembling the knowledge from pattern models using historical data and optimizing the interaction between them using real-time observations. It can dynamically estimate the priori pattern models combination in advance for the next time interval. The source demand data were collected by Smart Card system along one bus service route over one year. After correlation analysis, three temporal relevant pattern time series are generated, namely, the weekly, daily and hourly pattern time series. Then statistical pattern models are developed to capture different time series patterns. Finally, an amended IMM algorithm is applied to dynamically combine the pattern models estimations to output the final demand prediction. The proposed IMMPH model is validated by comparing with statistical methods and an artificial neural network based hybrid model. The results suggest that the IMMPH model provides a better forecast performance than its alternatives, including prediction accuracy, robustness, explanatory power and model complexity. The proposed approach can be potentially extended to other short-term time series forecast applications as well, such as traffic flow forecast.

Utilizing an Adaptive Grey Model for Short-Term Time Series Forecasting: A Case Study of Wafer-Level Packaging

The wafer-level packaging process is an important technology used in semiconductor manufacturing, and how to effectively control this manufacturing system is thus an important issue for packaging firms. One way to aid in this process is to use a forecasting tool. However, the number of observations collected in the early stages of this process is usually too few to use with traditional forecasting techniques, and thus inaccurate results are obtained. One potential solution to this problem is the use of grey system theory, with its feature of small dataset modeling. This study thus uses the AGM(1,1) grey model to solve the problem of forecasting in the pilot run stage of the packaging process. The experimental results show that the grey approach is an appropriate and effective forecasting tool for use with small datasets and that it can be applied to improve the wafer-level packaging process.

The Effects of Predictor Variables and Interval Partition on Fuzzy Time Series Forecasting

Fuzzy time series methods have been applied to social forecasting for over a decade; however, little research has been done to discuss the decision of an optimal fuzzy model for time series. In the paper, we evaluate the forecasting performance of three listed multivariate fuzzy models by comparing forecasting MSE of model. The data obtained from AEROM, Taiwan, includes Taiwan’s exports and foreign exchange rate for models’ test. The algorithm for predictive value of the models has three-stage computation procedure: First, calibrating time series correlation, deciding window base and interval partition; second, solving the static forecasting value of each model; third, comparing the dynamic parameter to impact of the forecasting error. The empirical results indicate that increasing predictor variables has no significant effect on predictive performance of the models; increasing length of interval would not improve the prediction performance of the models. Moreover, Fuzzy model is better for short-term time series forecasting. For forecasting purpose, Heuristic model has best forecasting performance among three fuzzy models. The findings of the paper represent a significant contribution to our understanding of the applicability of fuzzy models to predict.

Applying nonlinear generalized autoregressive conditional heteroscedasticity to compensate ANFIS outputs tuned by adaptive support vector regression

Volatility clustering degrades the efficiency and effectiveness of time series prediction and gives rise to large residual errors. This is because volatility clustering suggests a time series where successive disturbances, even if uncorrelated, are yet serially dependent. To overcome volatility clustering problems, an adaptive neuro-fuzzy inference system (ANFIS) is combined with a nonlinear generalized autoregressive conditional heteroscedasticity (NGARCH) model that is tuned by adaptive support vector regression (ASVR) so as to tackle the problem of time-varying conditional variance in residual errors. The proposed method significantly reduces large residual errors in forecasts because volatility clustering effects are regulated to trivial levels. Two experiments using real financial data series compare the proposed method and a number of well-known alternative methods. Results show that forecasting performance by the proposed method produces superior results, with good speed of computation. Goodness of fit of the proposed method is tested by Ljung–Box Q-test. It is concluded that the ANFIS/NGARCH composite model tuned by ASVR performs very well for improved predictive accuracy of irregular non-periodic short-term time series forecast and will be of interest to the science of statistical prediction of time series.