Forecasting Strategies Research Articles

Hybrid learning strategies for multivariate time series forecasting of network quality metrics

This work addresses the challenge of forecasting temporal metrics that characterize cellular traffic behavior. The ultimate goal is to provide network operators with a valuable tool for modeling mobile network traffic and optimizing connected resources. The idea is to estimate beforehand the temporal evolution of some Quality-of-Experience (QoE) and Quality-of-Service (QoS) metrics, which is helpful for accurately tuning the allocation of network resources. Remarkably, these metrics (expressed as time series) are typically correlated, and changes in one time series can affect others in a variety of ways and to different extents. For example, high network delay (a QoS-related metric) is associated with degradation in voice quality over time (a QoE-related metric). Accordingly, we address the problem of cellular traffic forecasting with correlated time series, proposing three innovative hybrid learning strategies designed by combining the advantages of two approaches: (i) a statistical approach, implemented through the Vector Autoregressive (VAR) model, which encodes each metric as a combination of past values of the same metric along with a combination of values of other related metrics, resulting in a multivariate structure; and (ii) an approach based on deep learning techniques (specifically, CNN, LSTM, and GRU) which operate on such a multivariate structure to perform the forecasting. The resulting performance demonstrates the benefits of the proposed hybrid schemes (VAR-CNN, VAR-LSTM, VAR-GRU) over their pure counterparts, with a significant reduction in forecasting errors. The network metrics were gathered in a real urban cellular environment, where the presence of exogenous factors (e.g., interferences, weather conditions, etc.) makes the forecasting assessment particularly challenging.

An open online simulation strategy for hydrological ensemble forecasting

Hydrological ensemble forecasting is crucial for flood forecasting. However, the centralized architecture of hydrological ensemble forecasting systems requires time- and labour-intensive downloads and the installation of executable models and methods. This usage pattern impedes the reusability of forecasting models and techniques. To address these limitations, we propose an open online simulation strategy with three components: model sharing and integration, data sharing and adaptation, and parameter optimization and recommendation. The model sharing and integration method helps researchers publish forecasting models as web services for online simulation and integration. A reusable data sharing and adaptation method is established for managing and processing data to meet model requirements. In addition, the optimization and recommendation methods are intended to assist researchers in optimizing and recommending model parameters online based on the characteristics of different research regions. Finally, a prototype system and case study are constructed to verify the strategy's feasibility and capability.

A novel deep learning-based evolutionary model with potential attention and memory decay-enhancement strategy for short-term wind power point-interval forecasting

Wind power generation plays a crucial role in promoting the transformation and advancement of the power industry and fostering sustainable development in society. However, wind power generation is susceptible to external factors, exhibiting significant volatility and randomness, which can adversely affect the stable operation of the power grid. The uncertainty associated with wind power generation can be effectively addressed through wind power prediction technology. While numerous wind power prediction methods have been developed, challenges remain, including imperfect data processing mechanisms and the optimization of model parameters, which hinder the effective utilization of wind power generation. To overcome these challenges, this research proposes a novel deep learning-based evolutionary model with a potential attention mechanism and memory decay-enhancement strategy for short-term wind power point-interval forecasting. The proposed model demonstrates high prediction stability and accuracy for both point values and intervals of short-term wind power, even under complex environmental and multi-seasonal conditions. The effectiveness of the proposed methods and strategies are validated using a measured dataset from the La Haute Borne wind farm in France. The results consistently show that, in complex environmental scenarios, the point prediction evaluation index of determination coefficient exceeds 90% and the interval prediction evaluation index of PI coverage probability exceeds 80%. Accurate short-term wind power point-interval forecasting contributes to enhancing the stable operation of the power system and improving the efficiency of wind energy utilization.

Comparative study of univariate and multivariate strategy for short-term forecasting of heat demand density: Exploring single and hybrid deep learning models

Accurate short-term forecasting of heating energy demand is needed for achieving optimal building energy management, cost savings, environmental sustainability, and responsible energy consumption. Furthermore, short-term heating energy prediction contributes to zero-energy building performance in cold climates. Given the critical importance of short-term forecasting in heating energy management, this study evaluated six prevalent deep-learning algorithms to predict energy load, including single and hybrid models. The overall best-performing predictors were hybrid models using Convolutional Neural Networks, regardless of whether they were multivariate or univariate. Nevertheless, while the multivariate models performed better in the first hour, the univariate models often were more accurate in the final 24 h. Thus, the best-performing predictor of the first timestep was a multivariate hybrid Convolutional Neural Network–Recurrent Neural Network model with a coefficient of determination (R²) of 0.98 and the lowest mean absolute error. Yet, the best-performing predictor of the final timestep was the univariate hybrid model Convolutional Neural Network–Long Short-Term Memory with an R² of 0.80. Also, the prediction accuracy of the best-performing multivariate hybrid models reduced faster per hour compared to the univariate models. These findings suggest that multivariate models may be better suited for early timestep predictions, while univariate models may be better suited for later time steps. Hence, combining the models can enhance accuracy at various timesteps for achieving high fidelity in forecasting and offering a comprehensive tool for energy management.

Seasonal predictability of the dominant surface ozone pattern over China linked to sea surface temperature

Mitigation surface ozone pollution becomes increasingly pivotal in improving China’s air quality. However, the impact of global sea surface temperature anomalies (SSTA) on the long-term predictability of China’s surface ozone remains challenging. In this study, we employ eigen techniques to effectively characterize dominant surface ozone patterns over China, and establish cross-correlations between the dominant patterns and global SSTA time series. Our findings reveal that China’s summer ozone pollution is strongly associated with crucial SSTA clusters linked to atmospheric circulations, i.e., the West Pacific Subtropical High and the Pacific-North American teleconnection pattern. For winter, ozone pollution is attributed to SSTA clusters related to the Southern Oscillation, the Madden-Julian Oscillation and others. We propose a multivariate regression model capable of predicting surface ozone patterns with a lead time of at least 3 months. Evaluation of our model using a testing dataset yields an R-value of around 0.5 between predicted and observed data, surpassing statistical significance threshold. This suggests the viability and potential applicability of our predictive model in surface ozone forecasting and mitigation strategies in China.

Forecasting value-at-risk and expected shortfall in emerging market: does forecast combination help?

PurposeThis paper investigates how various strategies for combining forecasts, both simple and optimised approaches, are compared with popular individual risk models in estimating value-at-risk (VaR) and expected shortfall (ES) in emerging market at alternative risk levels.Design/methodology/approachUsing the case study of the Vietnamese stock market, the author produced one-day-ahead VaR and ES forecast from seven individual risk models and ten alternative forecast combinations. Next, the author employed a battery of backtesting procedures and alternative loss functions to evaluate the global predictive accuracy of the different methods. Finally, the author investigated the relative performance over time of VaR and ES forecasts using fluctuation test.FindingsThe empirical results indicate that, although combined forecasts have reasonable predictive abilities, they are often outperformed by one individual risk model. Furthermore, the author showed that the complex combining methods with optimised weighting functions do not perform better than simple combining methods. The fluctuation test suggests that the poor performance of combined forecasts is mainly due to their inability to cope with periods of instability.Research limitations/implicationsThis study reveals the limitation of combining strategies in the one-day-ahead VaR and ES forecasts in emerging markets. A possible direction for further research is to investigate whether this finding holds for multi-day ahead forecasts. Moreover, the inferior performance of combined forecasts during periods of instability motivates further research on the combining strategies that take into account for potential structure breaks in the performance of individual risk models. A potential approach is to improve the individual risk models with macroeconomic variables using a mixed-data sampling approach.Originality/valueFirst, the authors contribute to the literature on the forecasting combinations for VaR and ES measures. Second, the author explored a wide range of alternative risk models to forecast both VaR and ES with recent data including periods of the COVID-19 pandemic. Although forecast combination strategies have been providing several good results in several fields, the literature of forecast combination in the VaR and ES context is surprisingly limited, especially for emerging market returns. To the best of the author’s knowledge, this is the first study investigating predictive power of combining methods for VaR and ES in an emerging market.

A Novel Sequence-to-Sequence-Based Deep Learning Model for Multistep Load Forecasting.

Load forecasting is critical to the task of energy management in power systems, for example, balancing supply and demand and minimizing energy transaction costs. There are many approaches used for load forecasting such as the support vector regression (SVR), the autoregressive integrated moving average (ARIMA), and neural networks, but most of these methods focus on single-step load forecasting, whereas multistep load forecasting can provide better insights for optimizing the energy resource allocation and assisting the decision-making process. In this work, a novel sequence-to-sequence (Seq2Seq)-based deep learning model based on a time series decomposition strategy for multistep load forecasting is proposed. The model consists of a series of basic blocks, each of which includes one encoder and two decoders; and all basic blocks are connected by residuals. In the inner of each basic block, the encoder is realized by temporal convolution network (TCN) for its benefit of parallel computing, and the decoder is implemented by long short-term memory (LSTM) neural network to predict and estimate time series. During the forecasting process, each basic block is forecasted individually. The final forecasted result is the aggregation of the predicted results in all basic blocks. Several cases within multiple real-world datasets are conducted to evaluate the performance of the proposed model. The results demonstrate that the proposed model achieves the best accuracy compared with several benchmark models.

Forecasting COVID-19 Pandemic – A scientometric Review of Methodologies Based on Mathematics, Statistics, and Machine Learning

Introduction: The COVID-19 pandemic is being regarded as a worldwide public health issue. The virus has disseminated to 228 nations, resulting in a staggering 772 million global infections and a significant death toll of 6.9 million. Since its initial occurrence in late 2019, many approaches have been employed to anticipate and project the future spread of COVID-19. This study provides a concentrated examination and concise evaluation of the forecasting methods utilised for predicting COVID-19. To begin with, A comprehensive scientometric analysis has been conducted using COVID-19 data obtained from the Scopus and Web of Science databases, utilising bibliometric research. Subsequently, a thorough examination and classification of the existing literature and utilised approaches has been conducted. First of its kind, this review paper analyses all kinds of methodologies used for COVID-19 forecasting including Mathematical, Statistical, Artificial Intelligence - Machine Learning, Ensembles, Transfer Learning and hybrid methods. Data has been collected regarding different COVID-19 characteristics that are being taken into account for prediction purposes, as well as the methodology used to develop the model. Additional statistical analysis has been conducted using existing literature to determine the patterns of COVID-19 forecasting in relation to the prevalence of methodologies, programming languages, and data sources. This review study may be valuable for researchers, specialists, and decision-makers concerned in administration of the Corona Virus pandemic. It can assist in developing enhanced forecasting models and strategies for pandemic management.

Daily Load Forecasting and Data-Driven Strategies for Steel Industry Based on Random Forest Modeling

Abstract As a large power consumer, the iron and steel industry urgently needs to improve productivity, reduce energy consumption, and save costs by revolutionizing energy management. In this study, we design a power demand management system for the iron and steel industry, and around the load management module, we propose a data-driven strategy based on daily load situational awareness and introduce the random forest model into daily load forecasting in the iron and steel industry. At the same time, the projection principle is applied to improve the traditional gray correlation similar day selection algorithm, and a combination method of daily load forecasting based on the gray projection improved random forest algorithm is proposed. The electric load data of the iron and steel industry in a specific region is utilized as an experimental sample to investigate the model’s forecasting performance and the impact of the data-driven strategy. The model for daily load forecasting in this paper has an average relative error of 1.18%, which is better than other models. The application of the data-driven strategy brought about 6.99% and 6.69% reductions in load demand and basic electricity cost. The data-driven strategy for the steel industry based on the Random Forest model can predict electric loads more accurately and reduce energy costs, as shown by the results.

The univariate model for long-term wind speed forecasting based on wavelet soft threshold denoising and improved Autoformer

Accurate wind speed prediction is crucial for effective wind power grid integration and energy dispatching. Recent research has explored the combination of decomposition algorithms with forecast models to form hybrid models, aiming to enhance wind speed prediction accuracy. However, these traditional decomposition techniques often lead to high time costs in practical applications, as they require new wind speed sequences to be appended to historical long sequences for decomposition before entering the forecast model. To overcome this challenge, this study introduces and improves the Autoformer model, applying it for the first time to long-term univariate wind speed forecasting. By incorporating decomposition technology as a sub-module of the forecast model, Autoformer not only solves the high time cost issue found in conventional hybrid models but also retains the benefits of decomposition technology in time series processing.Furthermore, in this paper, the decomposition module of Autoformer is replaced with the Mixture of Expert Decomposition Module (MOEDecomp) to better extract complex trend elements of wind speed series. Combined with the auto-correlation mechanism, sequential attention is paid to wind speed for extracting time dependencies in long series. Additionally, the Wavelet Soft Threshold Denoising (WSTD) algorithm is utilised for noise reduction in wind speed sequences. To evaluate the model's performance, two multi-step forecasting strategies were employed to predict wind speeds for the forthcoming 24, 48, 72, 96, and 120 h using four datasets. Experimental results demonstrate that the proposed model surpasses 15 comparative models in terms of prediction accuracy, generalization ability, and handling uncertain data.

Research on Smart Grid Forecasting Strategy Based on Artificial Intelligence

Research on Smart Grid Forecasting Strategy Based on Artificial Intelligence

ЧИННИКИ ТА ПЕРЕДУМОВИ ФОРМУВАННЯ НОВИХ ТИПОЛОГІЧНИХ ОЗНАК РЕГІОНУ СХОДУ УКРАЇНИ

The prerequisites and factors of the formation of a new type of region in the east of Ukraine have been studied. The principles and tasks of effective regional policy in modern conditions have been established. Typological differences make it possible to determine the patterns of functioning, formation and development of regions. Problems and tasks of the development of regions that are solved in the process of urban planning depend on the type and main features of the function and shape of the region. After the occupation, the state will receive two factors that will shape the problem of changing the territorial planning organization of the region and region. This is the border factor and the reintegration factor of the region that was under occupation for a long time. Trends in the development of urbanized territories and urban agglomerations strengthen the processes of formation of territorial settlement systems at the sub-regional, regional and sub-district levels. Generalized goals of regional policy determine the forecast strategy of the region's development. The structure of the regional forecast should include the following types: demographic, natural resource, economic. The general scheme of planning the territory of Ukraine aims to establish state priorities regarding rational types and modes of use of the country's territory, taking into account socio-economic needs, environmental limitations, resource opportunities and regional differences. The general scheme defines the prerequisites that contribute to the use of the territory of Ukraine. For the eastern region of Ukraine, there is a task of functional arrangement of the planning structure, taking into account the change in form. The main formative aspect is the border of the state. Objects, areas of military purpose, which form the supporting frame of the security belt, also become a derivative element. The urban planning analysis allows us to conclude that the border region will have a configuration with an impact zone of up to 50 km - the first zone of existential danger. Up to 200 km of the general strip of the border region, which will feel the influence of the border with the enemy state. In the first zone, the implementation of measures of a restrictive nature, structural transformation of communication routes and roads, pipeline infrastructure, etc., a general decrease in the population and network of settlements is expected. In the general strip, inter-village connections are formed, oriented in the direction of the center. A potential threat is the deindustrialization of the region, the decline of single settlements focused on the coal mining industry.

Comparing calibrated analog and dynamical ensemble solar forecasts

Ensemble modeling is a chief strategy for probabilistic forecasting. In weather forecasting, analog ensemble, which operates under the principle that weather patterns often repeat, and dynamical ensemble, which generates equally likely trajectories of future weather by perturbing the initial and boundary conditions, constitute the two most common approaches to making ensembles. That said, in the field of solar forecasting, nor is there any head-to-head comparison made thus far in regard to understanding the relative performance of these two competing approaches; this work seeks to fill the gap. Four years (2017–2020) of operational forecasts, at seven locations, from the European Centre for Medium-Range Weather Forecasts (ECMWF) are used, and both the raw and post-processed versions of the ensemble irradiance forecasts are verified in a fair and thorough fashion. Three classical post-processing methods, namely, Bayesian model averaging, nonhomogeneous Gaussian regression, and quantile regression, are applied to both the analog ensemble forecasts derived from ECMWF’s high-resolution control forecasts and dynamical ensemble forecasts from ECMWF’s Ensemble Prediction System. It is found that analog ensemble before post-processing possesses some advantage in terms of calibration over dynamical ensemble; their average reliability values are 0.6 W/m2 and 8.2 W/m2, respectively. However, dynamical ensemble after post-processing becomes generally more attractive, obtaining an average continuous ranked probability score of 49.0 W/m2, against 51.7 W/m2 for AnEn.

Oil Price Exposure and the Cross-Section of Stock Returns

Abstract We provide evidence that equity investors are slow to process information about how current oil price changes affect future earnings announcements. Stock prices respond to lagged quarterly oil price changes when firms start announcing earnings in the next quarter. A cross-sectional equity trading strategy that exploits this predictability yields an annualized Sharpe ratio of 0.50. Our oil-response forecast strategy earns especially high returns after large absolute oil price changes, in recessions or bear markets, and during peak earnings season. The predictability we document is consistent with limited attention, is not driven by risk factor exposure, and survives several robustness tests. (JEL G10, G11, G14, G40, Q41)

A generalized feature projection scheme for multi-step traffic forecasting

Exploiting spatial–temporal correlations has long been regarded as the cornerstone of traffic state prediction. Among existing techniques, temporal graph neural networks (TGNNs) have recently emerged as a prominent solution for modeling complex spatial–temporal traffic data correlations. Existing studies on TGNNs mainly focus on developing new building blocks to embed hidden correlations into a unified latent representation, which is mapped to predictions of distinct horizons. However, mapping the same latent features to distinct scalar predictions makes the gradient computation challenging for updating model parameters in the relevant directions. Besides, TGNNs are biased towards the shared temporal patterns while neglecting the complex dependencies within each data series, which can be captured to enrich latent features. To handle these problems jointly, we propose a novel feature projection scheme for the traffic prediction framework of TGNNs. The proposed projection scheme is based on spatial convolutions that first generate horizon-specific feature maps and then transform them into scalar predictions of the corresponding horizons. These horizon-specific feature maps establish interactions between the unified latent representation and the corresponding output values to bring the predictions closer to the true values. Besides, the proposed scheme also serves as a pattern modeling phase that enhances the expressivity of TGNNs by enriching latent features with data source-wise patterns of distinct time steps. Comprehensive experiments on two real-world traffic datasets demonstrate that the proposed scheme enhances the predictive performance and reduces the model parameters of TGNNs.

Improving the Forecasting of El Niño Amplitude Based on an Ensemble Forecast Strategy for Westerly Wind Bursts

Abstract The forecasting of El Niño amplitude remains uncertain, with false alarms or underestimations often occurring. It has been suggested that westerly wind bursts (WWBs) are crucial atmospheric stochastic forcing that affects the development, amplitude, and predictability of El Niño. Effectively capturing the influence of WWBs in El Niño forecasting systems to mitigate El Niño forecast uncertainties continues to be a widely discussed topic. In this work, two El Niño ensemble forecast scenarios incorporating WWBs were devised utilizing an intermediate coupled model capable of simulating the rational features of El Niño–Southern Oscillation and a refined WWB parameterization scheme. To start with, both the seasonal variations and the dependence on the oceanic background state were considered to improve the parameterization of WWBs. Furthermore, considering the short intrinsic predictability limit of WWBs with respect to the interannual scale of El Niño evolution, a WWB ensemble forecast strategy was designed to obtain their occurrence probability and the statistical features of their magnitude and central location. With this in mind, an ensemble forecast of El Niño events occurring during 1979–2021 based on a WWB ensemble forecast was established (termed WWB-based). For comparison, a conventional El Niño ensemble forecast based on initial condition (IC) perturbations was also conducted (termed IC-based). Results indicated that the WWB-based approach shows better performance in forecasting the El Niño amplitude than the IC-based one. The present approach suggests that an ensemble forecast with proper consideration of the scale interaction between the fast WWBs and interannual variations is more physically consistent. Significance Statement Westerly wind bursts (WWBs) are believed to influence the growth and amplitude of El Niño events. Unfortunately, the short intrinsic predictability limit of WWBs brings considerable uncertainty to the forecasting of El Niño amplitude. In this study, a refined parameterization scheme for WWBs, along with an ensemble forecast strategy designed to obtain the occurrence probability of WWBs and the statistical features of their magnitude and central location, was employed in an intermediate coupled model to produce an ensemble forecast for the El Niño events occurring during 1979–2021. The results showed that an ensemble forecast of El Niño based on WWB ensemble forecast can fully consider WWBs’ potential impact on the El Niño system and improve El Niño forecasts.

Application of forecasting strategies and techniques to natural gas consumption: A comprehensive review and comparative study

Accurate forecasting of natural gas consumption (NGC) plays an important role in energy supply, energy trading, economic effects and environmental sustainability. NGC forecasts can be used to adjust production and supply plans to improve gas efficiency and reduce carbon emissions and supply chain waste. This paper reviews the research progress on NGC in the past decade, analyzes the typical characteristics of different forecasting strategies, and highlights 163 studies in terms of the technical aspects of feature processing methods, data decomposition methods, forecasting models and optimization algorithms. It also systematically elaborates the application of statistical models, machine learning models, grey models, logistic regression and their combinations in predictive models. Bibliometric methods are also utilized to dissect research hotspots and summarize cutting-edge trends in the field. It is worth mentioning that in the terms of hybrid model structures, the application and performance of various model structures are described and evaluated. In this paper, the future development is discussed from spatiotemporal characteristics, studying reasonable data decomposition layers and fusion models, considering potential data privacy issues, and developing artificial intelligence-supporting models and interpretable frameworks. This paper is expected to provide a multi-technology reference for natural gas forecasting and help researchers to select and develop more accurate forecasting techniques and strategies.

A Review of Deep Transfer Learning Strategy for Energy Forecasting

Over the past decades, energy forecasting has attracted many researchers. The electrification of the modern world influences the necessity of electricity load, wind energy, and solar energy forecasting in power sectors. Energy demand increases with the increase in population. The energy has inherent characteristics like volatility and uncertainty. So, the design of accurate energy forecasting is a critical task. The electricity load, wind, and solar energy are important for maintaining the energy supply-demand equilibrium non-conventionally. Energy demand can be handled effectively using accurate load, wind, and solar energy forecasting. It helps to maintain a sustainable environment by meeting the energy requirements accurately. The limitation in the availability of sufficient data becomes a hindrance to achieving accurate energy forecasting. The transfer learning strategy supports overcoming the hindrance by transferring the knowledge from the models of similar domains where sufficient data is available for training. The present study focuses on the importance of energy forecasting, discusses the basics of transfer learning, and describes the significance of transfer learning in load forecasting, wind energy forecasting, and solar energy forecasting. It also explores the reviews of work done by various researchers in electricity load forecasting, wind energy forecasting, and solar energy forecasting. It explores how the researchers utilized the transfer learning concepts and overcame the limitations of designing accurate electricity load, wind energy, and solar energy forecasting models.

Generative AI in the Manufacturing Process: Theoretical Considerations

Abstract The paper aims to identify how digital transformation and Generative Artificial Intelligence (GAI), in particular, affect the manufacturing processes. Several dimensions of the Industry 4.0 field have been considered, such as the design of new products, workforce and skill optimisation, enhancing quality control, predictive maintenance, demand forecasting, and marketing strategy. The paper adopts qualitative research based on a critical review approach. It provides evidence of the GAI technology support in the mentioned areas. Appropriate use of emerging technology allows managers to transform manufacturing by optimising processes, improving product design, enhancing quality control, and contributing to overall efficiency and innovation in the industry. Simultaneously, GAI technologies facilitate predictive analytics to forecast and anticipate future demand, quality issues, and potential risks, improve a marketing strategy and identify market trends.

Out‐of‐sample volatility prediction: Rolling window, expanding window, or both?

AbstractEstimation windows, either rolling or expanding, are used for volatility forecasting. In this study, we propose a new approach relying on both estimation windows. Our method is based on how well these two windows performed in terms of prediction during a recent period of past time. We will continue to use whichever one has performed better in the past. Results show that our strategy significantly outperforms the individual and mean combination models. Whether the window is rolling or expanding, the relatively better performance is persistent. In other words, we document the existence of the momentum of predictability (MoP). A mean–variance investor can achieve the highest utility gains using our strategy for volatility forecasting. Moreover, the results pass a series of robustness tests.