Marlon Núñez

Real‐Time Prediction of the Occurrence and Intensity of the First Hours of >100 MeV Solar Energetic Proton Events

A new model for predicting the occurrence of >100 MeV solar energetic proton (SEP) events and the first hours of the >100 MeV integral proton flux is presented. This model uses a novel approach based on the lag correlation between strong positive derivatives of X-ray flux and proton flux. The new model has been validated with data from January 1994 to September 2013, obtaining a probability of detection of all >100 MeV SEP events of 80.85%, a false alarm ratio of 29.62%, and an average warning time of 1 h and 6 min. The model identifies the associated flare and active region. Currently, there is no other automatic empirical or physics-based system able to predict SEP events of energies in the interval of 100 MeV to ~430 MeV (lower GLE cutoff according to Clem and Dorman (2000)). This paper also proposes the combined use of the new prediction model and the existing one for predicting >10 MeV SEP events. The combined SEP prediction models have been developed to improve mitigation of adverse effects on near-Earth and interplanetary missions.

Self-Adaptive Induction of Regression Trees

A new algorithm for incremental construction of binary regression trees is presented. This algorithm, called SAIRT, adapts the induced model when facing data streams involving unknown dynamics, like gradual and abrupt function drift, changes in certain regions of the function, noise, and virtual drift. It also handles both symbolic and numeric attributes. The proposed algorithm can automatically adapt its internal parameters and model structure to obtain new patterns, depending on the current dynamics of the data stream. SAIRT can monitor the usefulness of nodes and can forget examples from selected regions, storing the remaining ones in local windows associated to the leaves of the tree. On these conditions, current regression methods need a careful configuration depending on the dynamics of the problem. Experimentation suggests that the proposed algorithm obtains better results than current algorithms when dealing with data streams that involve changes with different speeds, noise levels, sampling distribution of examples, and partial or complete changes of the underlying function.

Automatic discovery of rules for predicting network management events

In order to discover behavior patterns, current algorithms only analyze historical data in terms of performance data or fault events, ignoring the temporal correlation among different types of information, including the configuration changes. A method is presented that can discover recurrent patterns from multiple flows of events, such as alarms and configuration events, as well as discrete information, such as traffic and usage, taking into account static and dynamic information concerning observed objects and their environments. This method can filter out theoretically useless patterns, using a novel technique for detecting chaos in sequences of events. The prediction accuracy of the discovered patterns has been measured using objects with dynamic behavior controlled by known and complex differential equations. The proposed mining method has been used for discovering and predicting alarms in a computer network composed of several Internet servers taking into account the alarm and configuration events history, as well as static information about these servers.

Real-time prediction of the occurrence of GLE events

A tool for predicting the occurrence of Ground Level Enhancement (GLE) events using the UMASEP scheme [Nunez, 2011, 2015] is presented. This real-time tool, called HESPERIA UMASEP-500, is based on the detection of the magnetic connection, along which protons arrive in the near-Earth environment, by estimating the lag-correlation between the time derivatives of 1-minute soft X-ray flux (SXR) and 1-minute near-Earth proton fluxes observed by the GOES satellites. Unlike current GLE warning systems, this tool can predict GLE events before the detection by any neutron monitor (NM) station. The prediction performance measured for the period from 1986 to 2016 is presented for two consecutive periods, because of their notable difference in performance. For the 2000-2016 period, this prediction tool obtained a probability of detection (POD) of 53.8% (7 of 13 GLE events), a false alarm ratio (FAR) of 30.0%, and average warning times (AWT) of 8 min with respect to the first NM stations alert and 15 min to the GLE Alert Pluss warning. We have tested the model by replacing the GOES proton data with SOHO/EPHIN proton data, and the results are similar in terms of POD, FAR and AWT for the same period. The paper also presents a comparison with a GLE warning system. This project has received funding from the European Unions Horizon 2020 research and innovation programme under agreement No 637324.

On-Line learning of decision trees in problems with unknown dynamics

Learning systems need to face several problems: incrementality, tracking concept drift, robustness to noise and recurring contexts in order to operate continuously. A method for on-line induction of decision trees motivated by the above requirements is presented. It uses the following strategy: creating a delayed window in every node for applying forgetting mechanisms; automatic modification of the delayed window; and constructive induction for identifying recurring contexts. The default configuration of the proposed approach has shown to be globally efficient, reactive, robust and problem-independent, which is suitable for problems with unknown dynamics. Notable results have been obtained when noise and concept drift are present.

Learning Patterns of Behavior by Observing System Events

The proposed algorithm (BPL) induces behavior patterns from events taking into account characteristics of observed systems and their environment. The main strategy of this method consists on building summaries of the behaviour of a system as events arrive, and take these summaries as training examples. BPL constructs summaries with new features from events, like duration of current event values, repetitions of an event in a period of time, amongst others. This algorithm has been tested in learning faulty behavior of networks with the purpose of continuously predicting alarms.

Prediction of shock arrival times from CME and flare data

This paper presents the Shock ARrival Model (SARM) for predicting shock arrival times for distances from 0.72 AU to 8.7 AU by using coronal mass ejections (CME) and flare data. SARM is an aerodynamic drag model described by a differential equation that has been calibrated with a dataset of 120 shocks observed from 1997 to 2010 by minimizing the mean absolute error (MAE), normalized to 1 AU. SARM should be used with CME data (radial, earthward or plane-of-sky speeds), and flare data (peak flux, duration, and location). In the case of 1 AU, the MAE and the median of absolute errors were 7.0 h and 5.0 h respectively, using the available CME/flare data. The best results for 1 AU (an MAE of 5.8 h) were obtained using both CME data, either radial or cone-model-estimated speeds, and flare data. For the prediction of shock arrivals at distances from 0.72 AU to 8.7 AU, the normalized MAE and the median were 7.1 h and 5.1 h respectively, using the available CME/flare data. SARM was also calibrated to be used with CME data alone or flare data alone, obtaining normalized MAE errors of 8.9 h and 8.6 h respectively for all shock events. The model verification was carried out with an additional dataset of 20 shocks observed from 2010 to 2012 with radial CME speeds to compare SARM with the empirical ESA model [Gopalswamy et al., 2005a] and the numerical MHD-based ENLIL model [Odstrcil et al., 2004]. The results show that the ENLILs MAE was lower than the SARMs MAE, which was lower than the ESAs MAE. The SARMs best results were obtained when both flare and true CME speeds were used.

On forecasting the onset of Solar Proton Events

A major problem for predicting the onset of Solar Proton Events is the detection of the magnetic connection between the flare and the earth. If there is a magnetic connection, the particles accelerated by a large solar event may impact the earth and produce the onset of a solar energetic proton event. Current physical models cannot predict the onset of a SPE mainly because of the chaotic conditions within the IMF structure. Kiplinger (1995) reported a high correlation between the existence of 10 MeV protons at Earth and a characteristic pattern of X-ray spectral evolution for several associated flares. We propose a practical approach that tries to detect the time intervals of this correlation. Our assumption is that a high correlation betwewn X-ray and protons at Earth is an important symptom of a magnetic connection and may help to prevent Solar Proton Events.

HESPERIA Forecasting Tools: Real-Time and Post-Event

Within the HESPERIA Horizon 2020 project, two novel real-time tools to predict Solar Energetic Particle (SEP) events were developed. The HESPERIA UMASEP-500 tool makes real-time predictions using a lag-correlation between the soft X-ray (SXR) flux and high-energy differential proton fluxes of the GOES satellite network. We found that the use of proton data alone allowed this tool to make predictions before any Neutron Monitor (NM) station’s alert. The performance of this tool for predicting Ground Level Enhancement (GLE) events for the period 2000–2016 may be summarized as follows: the probability of detection (POD) was 53.8%, the false alarm ratio (FAR) was 30%, and the average warning time (AWT) to the first NM station’s alert was 8 min. The developed HESPERIA REleASE tool makes real-time predictions of the proton flux-time profiles of 30–50 MeV protons at L1 and is based on electron intensity measurements of energies from 0.25 to 1 MeV and their intensity changes. The performance was tested by using all historic ACE/EPAM and SOHO/EPHIN data from 2009 until 2016 and has shown that the forecast tools have a low FAR (∼30%) and a high POD (63%). Furthermore, two methods using historical data were explored for predicting SEP events and compared. The UMASEP-10mw tool was developed for predicting >10 MeV SEP events using microwave data. The time derivative of the soft X-rays (SXR) was replaced by the microwave flux density. It was found that the use of SXRs and microwave data produced the same POD (∼78%) with the most notable difference being that the use of microwave data does not yield any false alarm. Furthermore, a study was carried out on the possibility for the microwave emissions to be used to predict the spectral hardness of the SEP event and important results were deduced.

Incorporating Prediction Facilities to Autonomous Systems

In addition to current architecture levels (deliberative, executive and reactive), a new level is emerging due to recent advances in temporal patterns discovery. This new architecture level, the preventative level, prepares the whole system for predicted problems with the purpose of reducing potential risks. The performance of execution tasks, as well as deliberative plans, may be improved by providing autonomous systems with event prediction facilities. Thus, an autonomous system would prepare some preventative or preparatory actions to enable recoveries in the future. This paper summarizes the main functions of this four-level architecture and presents a testing procedure in the field of virtual robotics.