Forecast Window Research Articles

Min-max game based energy management strategy for fuel cell/supercapacitor hybrid electric vehicles

The fuel cell hybrid electric vehicles have received widespread attention in recent years due to their zero-emission and long mileage. However, the durability of the fuel cell declines rapidly under frequent start and stop and power fluctuation. The fuel efficiency of the power system is affected by the power allocation strategy. For low hydrogen consumption and fuel cell durability, an energy management strategy based on the game theory is proposed in this paper, which considers the controller and the future driving condition as both sides of the game to enhance the operational stability in complex driving conditions. The interval forecast for driving conditions is realized by introducing the Markov chain method. To obtain the optimal power allocation strategy in the speed forecast window, it is reasonable to integrate the co-evolutionary algorithm with the game model to leverage its superior ability in solving the constrained min-max problem. In order to verify the robustness and the economic performance of this method, the comparison with other methods is conducted in the typical urban, highway and mixed driving conditions. What is more, the fuel cell degradation and operational stability of four energy management strategies are evaluated by the fuel cell degradation model. The results show that the proposed method can reduce hydrogen consumption by more than 6.82% and avoid fuel cell degradation caused by frequent start and stop and power fluctuation compared with other real-time strategies.

Application of Data Fusion Techniques to Improve Air Quality Forecast: A Case Study in the Northern Italy

Deterministic air quality forecasting models play a key role for regional and local authorities, being key tools to ensure that timely information about actual or near future exceedances of pollutant threshold values are provided to the public, as stated by the EU directive (2008/50/EC). One of the main problems of these models is that they usually underestimate some important pollutants, like PM10, especially in high-concentration areas. For this reason, the forecast of critical episodes (i.e., exceedance of 50 μ g/m 3 for PM10 concentration daily threshold) has low accuracy. To overcome this issue, several computationally fast techniques have been implemented in the last decade. In this work, two computational fast techniques are introduced, implemented and evaluated. The techniques are based on the off-line correction of the chemical transport model output in the forecasting window, estimated by means of the measurement data up to the beginning of the forecast. In particular, the techniques are based on the estimation of the correction performed as a linear combination of the corrections computed for the days when the measurements are available. The resulting system has been applied to the Lombardy region case (Northern Italy) for daily PM10 forecasting with good results.

Short-term power load forecasting using integrated methods based on long short-term memory

The development of power system informatization, the massive access of distributed power supply and electric vehicles have increased the complexity of power consumption in the distribution network, which puts forward higher requirements for the accuracy and stability of load forecasting. In this paper, an integrated network architecture which consists of the self-organized mapping, chaotic time series, intelligent optimization algorithm and long short-term memory (LSTM) is proposed to extend the load forecasting length, decrease artificial debugging, and improve the prediction precision for the short-term power load forecasting. Compared with LSTM prediction, the algorithm in this paper improves the prediction accuracy by 61.87% in terms of root mean square error (RMSE), and reduces the prediction error by 50% in the 40-fold forecast window under some circumstances.

Generation Rejection Scheme Based-on a Combinational Rotor Angle Trajectory Prediction

This paper presents a response-based generating rejection scheme (GRS) based on an angular stability prediction logic to initiate the outage of accelerated generating units while saving the rest of generating units from the loss of synchronism. First trigonometric, polynomial, and hybrid models of rotor angle trajectory based on the reasonable assumptions are proofed. Then, by taking these models in the prediction step, through the maximum use of measured data based on defining the forecast horizon (FH) and data window with incremental length, the stability/instability of generating units is separately predicted. Next, the status of tripping signal based on a combinational logic of the output results of the angular stability prediction method is specified. In the developed logic, if at least two models of the three designated models yield the same response about the unit stability status, the trip signal is accordingly fired or blocked. The proposed method is examined on the one machine infinite bus and the WSCC standard test bed under different operation and fault scenarios. The obtained results demonstrate that beside simplicity, low computational burden, and very short processing time, the proposed combinatorial method outperforms the existing ones working with individual prediction models.

A dynamic neural network model for predicting risk of Zika in real time

BackgroundIn 2015, the Zika virus spread from Brazil throughout the Americas, posing an unprecedented challenge to the public health community. During the epidemic, international public health officials lacked reliable predictions of the outbreak’s expected geographic scale and prevalence of cases, and were therefore unable to plan and allocate surveillance resources in a timely and effective manner.MethodsIn this work, we present a dynamic neural network model to predict the geographic spread of outbreaks in real time. The modeling framework is flexible in three main dimensions (i) selection of the chosen risk indicator, i.e., case counts or incidence rate; (ii) risk classification scheme, which defines the high-risk group based on a relative or absolute threshold; and (iii) prediction forecast window (1 up to 12 weeks). The proposed model can be applied dynamically throughout the course of an outbreak to identify the regions expected to be at greatest risk in the future.ResultsThe model is applied to the recent Zika epidemic in the Americas at a weekly temporal resolution and country spatial resolution, using epidemiological data, passenger air travel volumes, and vector habitat suitability, socioeconomic, and population data for all affected countries and territories in the Americas. The model performance is quantitatively evaluated based on the predictive accuracy of the model. We show that the model can accurately predict the geographic expansion of Zika in the Americas with the overall average accuracy remaining above 85% even for prediction windows of up to 12 weeks.ConclusionsSensitivity analysis illustrated the model performance to be robust across a range of features. Critically, the model performed consistently well at various stages throughout the course of the outbreak, indicating its potential value at any time during an epidemic. The predictive capability was superior for shorter forecast windows and geographically isolated locations that are predominantly connected via air travel. The highly flexible nature of the proposed modeling framework enables policy makers to develop and plan vector control programs and case surveillance strategies which can be tailored to a range of objectives and resource constraints.

Testing for Changes in Forecasting Performance

We consider the issue of forecast failure (or breakdown) and propose methods to assess retrospectively whether a given forecasting model provides forecasts which show evidence of changes with respect to some loss function. We adapt the classical structural change tests to the forecast failure context. First, we recommend that all tests should be carried with a fixed scheme to have best power. This ensures a maximum difference between the fitted in and out-of-sample means of the losses and avoids contamination issues under the rolling and recursive schemes. With a fixed scheme, Giacomini and Rossi’s (GR) test is simply a Wald test for a one-time change in the mean of the total (the in-sample plus out-of-sample) losses at a known break date, say m, the value that separates the in and out-of-sample periods. To alleviate this problem, we consider a variety of tests: maximizing the GR test over values of m within a prespecified range; a Double sup-Wald (DSW) test which for each m performs a sup-Wald test for a change in the mean of the out-of-sample losses and takes the maximum of such tests over some range; we also propose to work directly with the total loss series to define the Total Loss sup-Wald and Total Loss UDmax (TLUD) tests. Using theoretical analyses and simulations, we show that with forecasting models potentially involving lagged dependent variables, the only tests having a monotonic power function for all data-generating processes considered are the DSW and TLUD tests, constructed with a fixed forecasting window scheme. Some explanations are provided and empirical applications illustrate the relevance of our findings in practice. Supplementary materials for this article are available online.

Internalizing the Harm of Privacy Breaches: Do Firms Have an Incentive to Improve Data Protection? An Event Study

Big data, combined with machine learning and artificial intelligence, is expected to increase the bulk of personal data in the hands of private and public entities, presenting a unique challenge to policymakers. Therefore, the question of how to incentivise firms to protect personal information is more relevant today than ever before. This study attempts to associate privacy breach announcements with the drop in share price using an event study. The results show that firms suffer a significant and immediate share price depreciation over a short window, and this is sustained over the entire forecast window after the event. Furthermore, this effect is greater when a larger number of customers are affected and the information exposed is financially sensitive. This represents a departure from previous studies, which found only a mild negative and temporary market reaction following a breach. These findings, using data from 2014-2018, will encourage policymakers to consider strict disclosure policies, thereby encouraging firms to invest in best data security practices.

A simple predictive model for the eddy propagation trajectory in the northern South China Sea

Abstract. A novel predictive model was built for eddy propagation trajectory using the multiple linear regression method. This simple model relates various oceanic parameters to eddy propagation position changes in the northern South China Sea (NSCS). These oceanic parameters mainly represent the effects of β and mean flow advection on the eddy propagation. The performance of the proposed model has been examined in the NSCS based on five years of satellite altimeter data and demonstrates its significant forecasting skills over a 4-week forecast window compared to the traditional persistence method. It was also found that the model forecasting accuracy is sensitive to eddy polarity and the forecast season.

Finite Sample Forecast Properties and Window Length Under Breaks in Cointegrated Systems

Finite Sample Forecast Properties and Window Length Under Breaks in Cointegrated Systems

Watershed-scale retrospective drought analysis and seasonal forecasting using multi-layer, high-resolution simulated soil moisture for Southeastern U.S

Drought assessment at local scales needs a reliable framework capturing both local landscape processes and the watershed-scale hydrological responses. Any such tool, implementable in a near real-time, semi-automated framework, is largely unavailable for the South Atlantic-Gulf (SAG) region in the Southeastern US (SEUS). In this study, we evaluate a drought monitoring and forecasting approach using multi-layer, high resolution, simulated soil moisture for 50 watersheds in the SEUS. Soil and Water Assessment Tool (SWAT) is integrated with meteorological drivers (precipitation and temperature) from Climate Forecast System Reanalysis data (CFSR) for retrospective simulations (January 1982–December 2013), and Climate forecast system model version-2 (CFSv2) models to obtain the near real-time estimates (January 2014 through mid-March 2017) and 9-month lead forecasts (mid-March through December 2017) of the hydrologic variables at 12-digit Hydrologic Unit Code (HUC-12) resolution. Drought assessment is carried out by combining the drought severity estimates of the weekly percentiles of the surface and total column (TC) soil moisture, aggregated to 4-week and 8-week respectively, following a different severity classification scheme for each layer. Several drought indices and observed drought maps from the U.S. Drought Monitor (USDM) are used to compare the agreement (or disagreement) of the proposed approach with the observed drought conditions in the region. The results show promising application of the proposed approach for near real-time estimation as the drought estimates show high (∼80–90%) Index of Agreement (IOA) with the Palmer Drought Severity Index (PDSI) for all drought categories (mild-exceptional). The retrospective assessment shows that TC soil moisture percentiles have high correlation (>0.7) with long-term drought indices. The surface soil moisture percentiles show high correlation (∼0.6) with Palmer Z Index (ZNDX) and 1-month Standardized Precipitation Index (SPI-1), while longer aggregations increase the association with the long-term drought indices. While the SWAT-CFSv2 based drought estimation is useful in near real-time mode, higher disagreement between the forecasted drought maps and the observed drought severity from the USDM is noted for a forecast window of greater than ∼ 4–6 weeks.

Short-term water demand forecasting using machine learning techniques

Abstract Nowadays, a large number of water utilities still manage their operation on the instant water demand of the network, meaning that the use of the equipment is conditioned by the immediate water necessity. The water reservoirs of the networks are filled using pumps that start working when the water level reaches a specified minimum, stopping when it reaches a maximum level. Shifting the focus to water management based on future demand allows use of the equipment when energy is cheaper, taking advantage of the electricity tariff in action, thus bringing significant financial savings over time. Short-term water demand forecasting is a crucial step to support decision making regarding the equipment operation management. For this purpose, forecasting methodologies are analyzed and implemented. Several machine learning methods, such as neural networks, random forests, support vector machines and k-nearest neighbors, are evaluated using real data from two Portuguese water utilities. Moreover, the influence of factors such as weather, seasonality, amount of data used in training and forecast window is also analysed. A weighted parallel strategy that gathers the advantages of the different machine learning techniques is suggested. The results are validated and compared with those achieved by autoregressive integrated moving average (ARIMA) also using benchmarks.

Neighbourhood\xa0level real-time forecasting of dengue cases in tropical urban Singapore

BackgroundDengue, a vector-borne infectious disease caused by the dengue virus, has spread through tropical and subtropical regions of the world. All four serotypes of dengue viruses are endemic in the equatorial city state of Singapore, and frequent localised outbreaks occur, sometimes leading to national epidemics. Vector control remains the primary and most effective measure for dengue control and prevention. The objective of this study is to develop a novel framework for producing a spatio-temporal dengue forecast at a neighbourhood level spatial resolution that can be routinely used by Singapore’s government agencies for planning of vector control for best efficiency.MethodsThe forecasting algorithm uses a mixture of purely spatial, purely temporal and spatio-temporal data to derive dynamic risk maps for dengue transmission. LASSO-based regression was used for the prediction models and separate sub-models were constructed for each forecast window. Data were divided into training and testing sets for out-of-sample validation. Neighbourhoods were categorised as high or low risk based on the forecast number of cases within the cell. The predictive accuracy of the categorisation was measured.ResultsClose concordance between the projections and the eventual incidence of dengue were observed. The average Matthew’s correlation coefficient for a classification of the upper risk decile (operational capacity) is similar to the predictive performance at the optimal 30% cut-off. The quality of the spatial predictive algorithm as a classifier shows areas under the curve at all forecast windows being above 0.75 and above 0.80 within the next month.ConclusionsSpatially resolved forecasts of geographically structured diseases like dengue can be obtained at a neighbourhood level in highly urban environments at a precision that is suitable for guiding control efforts. The same method can be adapted to other urban and even rural areas, with appropriate adjustment to the grid size and shape.

Using Machine Learning Methods to Forecast if Solar Flares Will Be Associated with CMEs and SEPs

Abstract Among the eruptive activity phenomena observed on the Sun, those that threaten human technology the most are flares with associated coronal mass ejections (CMEs) and solar energetic particles (SEPs). Flares with associated CMEs and SEPs are produced by magnetohydrodynamical processes in magnetically active regions (ARs) on the Sun. However, these ARs do not only produce flares with associated CMEs and SEPs, they also lead to flares and CMEs, which are not associated with any other event. In an attempt to distinguish flares with associated CMEs and SEPs from flares and CMEs, which are unassociated with any other event, we investigate the performances of two machine learning algorithms. To achieve this objective, we employ support vector machines (SVMs) and multilayer perceptrons (MLPs) using data from the Space Weather Database of Notification, Knowledge, Information of NASA Space Weather Center, the Geostationary Operational Environmental Satellite, and the Space-Weather Heliospheric and Magnetic Imager Active Region Patches. We show that True Skill Statistics (TSS) and Heidke Skill Scores (HSS) calculated for SVMs are slightly better than those from the MLPs. We also show that the forecasting time frame of 96 hr provides the best results in predicting if a flare will be associated with CMEs and SEPs (TSS = 0.92 ± 0.09 and HSS = 0.92 ± 0.08). Additionally, we obtain the maximum TSS and HSS values of 0.91 ± 0.06 for predicting that a flare will not be associated with CMEs and SEPs for the 36 hr forecast window, while the 108 hr forecast window gives the maximum TSS and HSS values for predicting that CMEs will not be accompanying any events (TSS = HSS = 0.98 ± 0.02).

Spatio-temporal hazard estimation in San Miguel volcano, El Salvador

Despite being one of the most active volcanoes in El Salvador, San Miguel is surrounded by important population nuclei and infrastructures. We used existing historical records of past eruptive activity, available geological knowledge and monitoring data gathered over the past sixteen years to conduct the first comprehensive, long-term hazard assessment of this volcano, aimed at helping to reduce the potential risk it poses. We applied probabilistic methods (QVAST and HASSET) specifically designed for volcanic hazard assessment to conduct two hazard analyses, one with a forecasting time window of two years using information on volcanic activity over the past 430years (historical period), and another with a forecasting window of six months, with information from the past 16years (monitoring period). In both cases we used the same susceptibility map constructed using the most relevant structural elements, which indicates the spatial probabilities of hosting a new vent on San Miguel volcano. Then, we calculated the most likely eruptive scenarios for each analysis and their corresponding forecasting windows, which revealed that the most hazardous areas correspond to Conacastal canton and hamlets of San Jorge, San Miguel and San Rafael Oriente – all lie within 5km of the central crater on the northern and southern flanks. This methodology is easy to update, and results can be improved by incorporating new information. This long-term hazard assessment is a significant contribution that will enable local authorities to implement more rational territorial planning and to design better emergency plans for coping with future volcanic crises.

Intelligent energy management strategy based on hierarchical approximate global optimization for plug-in fuel cell hybrid electric vehicles

The energy management strategy (EMS) is a key to reduce the equivalent hydrogen consumption and slow down fuel cell performance degradation of the plug-in fuel cell hybrid electric vehicles. Global optimal EMS based on the whole trip information can achieve the minimum hydrogen consumption, but it is difficult to apply in real driving. This paper tries to solve this problem with a novel hierarchical EMS proposed to realize the real-time application and approximate global optimization. The long-term average speed in each future trip segment is predicted by KNN, and the short-term speed series is predicted by a new model averaging method. The approximate global optimization is realized by introducing hierarchical reinforcement learning (HRL), and the strategy within the speed forecast window is optimized by introducing upper confidence tree search (UCTS). The vehicle speed prediction and the proposed EMS have been verified using the collected real driving cycles. The results show that the proposed strategy can adapt to driving style changes through self-learning. Compared with the widely used rule-based strategy, it can evidently reduce hydrogen consumption by 6.14% and fuel cell start-stop times by 21.7% on average to suppress the aging of fuel cell. Moreover, its computation time is less than 0.447 s at each step, and combined with rolling optimization, it can be used for real-time application.

Forecasting gold futures market volatility using macroeconomic variables in the United States

The U.S. gold futures market has recently attracted significant attention and the gold volatility is closely linked to macroeconomics. As such, the question is how to analyze the impact of various macroeconomic variables on gold. We use the GARCH-MIDAS (mixed data sampling) model to investigate whether macroeconomic variables can improve the predictions on the volatility structure of U.S. gold futures. Our empirical results reveal that macroeconomic variables have a significant influence on gold volatility, especially during and after the global financial crisis, indicating macroeconomic variables are driving factors of the long-term volatility on the U.S. gold futures market. Additionally, we use principal component analysis to obtain key information on different macroeconomic variables and further investigate their joint effects on the volatility of gold futures, finding that the first and second principal components are good proxies of macroeconomic variables. Our results show that principal components improve forecast accuracy, as do macrovariables, which are robust to various forecast rolling window schemes.

Forecasting Solar Flares Using Magnetogram-based Predictors and Machine Learning

We propose a forecasting approach for solar flares based on data from Solar Cycle 24, taken by the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO) mission. In particular, we use the Space-weather HMI Active Region Patches (SHARP) product that facilitates cut-out magnetograms of solar active regions (AR) in the Sun in near-real-time (NRT), taken over a five-year interval (2012 - 2016). Our approach utilizes a set of thirteen predictors, which are not included in the SHARP metadata, extracted from line-of-sight and vector photospheric magnetograms. We exploit several Machine Learning (ML) and Conventional Statistics techniques to predict flares of peak magnitude >M1 and >C1, within a 24 h forecast window. The ML methods used are multi-layer perceptrons (MLP), support vector machines (SVM) and random forests (RF). We conclude that random forests could be the prediction technique of choice for our sample, with the second best method being multi-layer perceptrons, subject to an entropy objective function. A Monte Carlo simulation showed that the best performing method gives accuracy ACC=0.93(0.00), true skill statistic TSS=0.74(0.02) and Heidke skill score HSS=0.49(0.01) for >M1 flare prediction with probability threshold 15% and ACC=0.84(0.00), TSS=0.60(0.01) and HSS=0.59(0.01) for >C1 flare prediction with probability threshold 35%.

Evaluation of the Impact of Data Uncertainty on the Prediction of Physiological Patient Deterioration

In this paper, we investigate the impact of data uncertainty on the prediction of a vital sign of abnormalities. We analyze the evolution of mean arterial blood pressure (MAP) and heart rate (HR) vital signs time series over a given forecasting window, whose values are classified according to a model of normality that models patient’s physiological deterioration. To this end, we developed a generic prediction framework that enables the deployment of different methods for data preparation, training, prediction, and classification so as to perform a set of experiments. To address our purpose, we selected a general regression neural network model (to predict vital sign trends) and a support vector machine (to recognize abnormalities) as the appropriate methods for a baseline analysis. However, other algorithm choices could be tailored in the defined prediction framework. We extract from the multiparameter intelligent monitoring in intensive care II waveform database a set of 302, 940 vital sign samples (organized as 459 records) to define training and testing data sets. From our experiments, we obtained evidence demonstrating the incidence of null values in the MAP/HR input time series on forecasting tasks. This was performed under different conditions in terms of null values in the observation window, and varying other factors like labeling methods and forecast window sizes. In our experiments, we concentrate on analyzing the extent of prediction performance under such conditions and identify that a tolerable level of null values is around 25% where it is possible to still derive forecast windows with an accuracy target $\geq 0.9$ for both MAP and HR vital signs. The presented evaluation contributes to acknowledge the importance of uncertainty conditions on the prediction performance of vital signs and paves the way for future extensions involving other prediction approaches.

Comparing in- and out-of-sample approaches to variance decomposition-based estimates of network connectedness an application to the Italian banking system

We use methods that exploit variance decompositions from standard (rolling window) estimates of VAR(p) models to obtain estimates of network connectedness and perform an empirical comparison of the results derived under two alternative approaches: To base the decompositions on in-sample forecast errors vs. out-of-sample forecast errors derived from a separation between estimation and forecast evaluation window. Using the intraday realized variance of bank stock return, we derive novel empirical results on the systemic risk of the Italian banking system, whose network connectedness turns out to be generally high, increasing over time especially during the Great Financial and the European sovereign debt crises. However, whether a few net exporters of systemic risk may be reliably estimated turns out to depend on the methodology adopted when computing rolling window variance decompositions.

Lessons from near real-time forecasting of irregular leadership changes

Since 2014, we have been producing regular six-month forecasts of the probability of irregular leadership changes – coups, rebellions, protests that result in state leader changes – for most countries in the world for the Political Instability Task Force (PITF). During 2015, we issued new forecasts each month, with a delay as short as five days and no longer than two weeks into each six-month forecasting window. This article describes the approach we use to generate our forecasts and presents several examples of how we present forecasts. The forecasts are derived from a statistical ensemble of seven thematic models, each based on a split-population duration model that aims to capture a specific argument or related set of covariates. This approach is modular in that thematic models can be swapped out or new models integrated, and it lessens the need for generalist ‘kitchen sink’ models. Together, the models achieve high out-of-sample accuracy. Based on our experience, we draw conclusions about the practical, policy, and scientific aspects of this and similar undertakings. These include thoughts on how to evaluate and present forecasts, the potential role of ensembles in model comparison, the role of ensembles and prediction in causal research, and considerations for future efforts in forecasting and predictive modeling.