Prediction Of Extreme Values Research Articles

TXtreme: transformer-based extreme value prediction framework for time series forecasting

Time Series Forecasting (TSF) is crucial in various real-world applications such as climate forecasting and electricity demand prediction. Unlike traditional datasets, time series data points are influenced by their past values, necessitating specialized techniques to model these sequential dependencies specifically addressing non-linear patterns, abrupt changes, and outliers. The latest advancements have significantly enhanced TSF using machine learning and other methods. However, forecasting extreme events remains challenging. Extreme values, although rare, have significant real-world impacts such as heavy rainfall, fluctuations in electricity demand, and traffic surges. This paper proposes a TXtreme framework that uses Long-Short memory network, feed-forward neural network, and transformer to improve time series forecasting under extreme values. The model also uses statistical methods to explain the distribution of time series values. Extensive experiments are conducted using datasets from different domains to show the robustness of the proposed methodology. Results, derived by testing TXtreme on five datasets of different domain, indicate that TXtreme significantly outperforms state-of-the-art methods in time series forecasting, with improvements of 5–25% in root means squared error or mean absolute error. The proposed framework enhances TSF capabilities and ensures better generalization ability in extreme event forecasting, potentially leading to improved decision-making in critical applications.

Bike Sharing Demand Forecasting Based on the Informer Model

Advancements in machine learning and artificial intelligence have been instrumental in enhancing demand forecasting for shared transportation modes, significantly benefiting urban traffic management and promoting a balanced supply-demand dynamic. This study harnesses the deep learning Informer model to forecast bike-sharing demand, integrating multiple factors to provide an empirical foundation for transport scheduling and redeployment, thereby contributing to the advancement of the field. Our experimental analysis reveals that the Informer model outperforms LSTM in predictive accuracy, as evidenced by improved R² scores and MSE values, reflecting gains in training efficiency and predictive precision. Despite these advances, predictions for extreme demand values, particularly during peak usage periods, indicate potential areas for refinement. Future research directions are oriented towards optimizing model accuracy by leveraging multimodal data for a comprehensive analysis and enhancement of traffic management strategies. This approach aims not only to better predict demand and alleviate congestion but also to deliver more optimized solutions for businesses and convenience for users. Furthermore, the model's application extends beyond bike-sharing to other forms of shared mobility, such as electric scooters, potentially aiding in reducing carbon emissions and furthering the development of the sharing economy and environmental sustainability.

Prediction of ultimate tensions in mooring lines for a floating offshore wind turbine considering extreme gusts

The design of mooring line parameter for a floating offshore wind turbine (FOWT) should consider the ultimate limit state. Typhoons should be considered as a threatening condition for mooring design, and the characteristics of transient changing wind should be studied. Currently, few extreme value predictions consider gust features, and extreme mooring tensions may be underestimated in practical FOWT design. We conducted the simulations of the 5 MW wind turbine from the National Renewable Energy Laboratory (NREL) in the environmental conditions of extreme gusts, waves, and currents. Extreme gusts were found to cause dramatic increases in mooring tensions under 100-year typhoon conditions. The average conditional exceedance rate (ACER) method is used to predict extreme mooring tensions. The impacts of simulation duration and sample size on the prediction results are then discussed. The results show that in extreme gust conditions, the tail data of the average conditional exceedance rate of mooring tension presents a different extrapolation trend from stable wind conditions. Short-term predicted values in gust conditions are 14–20% larger than those in stable wind conditions. The research shows the necessity of considering threatening gust conditions in the ultimate limit state design of mooring system of FOWT.

Short-term extreme value prediction for the structural responses of the IEA 15 MW offshore wind turbine under extreme environmental conditions

As offshore wind turbines enter deep seas, large-scale models ranging from 15 MW to 20 MW are becoming the mainstream choice for the future market. The dynamic responses of 15 MW wind turbines under extreme environmental conditions should be considered in the design process. With increasing structure sizes, the mooring systems and tower base structures play essential roles in maintaining position and supporting wind turbines under harsh environmental conditions. Accurate estimation of the extreme mooring loads and dynamic responses at the tower base is crucial, particularly for the ultimate limit state (ULS) analysis. The dynamic responses of the offshore wind turbine are calculated by Open Fast and validated with other numerical tools. For the safety of deep-sea structures, it is necessary to consider the extreme environmental conditions with a 100-year return period, which is beyond the current standards. An environmental contour-based approach is proposed to select the short-term extreme conditions, enabling approximate prediction of the extreme values of the dynamic responses during the preliminary design stage. The Gumbel method and the ACER method are used to predict the short-term extreme dynamic responses. Comparative evaluations against the relevant standards for floating wind turbines are conducted to assess structural safety.

Evaluation of Extreme Value Predictions for Unsteady Flow Distortion of Aero-Engine Intakes

Abstract Unsteady flow distortion is of interest for the development of air-breathing propulsion systems. Stochastic fluctuations can generate incompatibilities between intakes and aero-engines. Observing the extreme flow distortion events during experimental testing is not guaranteed and statistical models such as extreme value theory (EVT) can be used to estimate the occurrence and magnitude of the fluctuations. However, the current industry standard does not provide guidance on how to apply these methods to obtain useful predictions. This work proposes a systematic process to assess the required number of observations for obtaining statistical convergence of the EVT predictions. This is achieved through shuffling of the data samples and relies on the availability of a sufficiently large initial dataset. This can be adopted by gas turbine engineers to evaluate the data recording requirements and to potentially reduce costs associated with experimental programs.

Extreme Value Prediction of Traffic Loads Using the Average Conditional Exceedance Rate Method

An efficient prediction of the extreme value of traffic loads is crucial for the structural design, reliability evaluation, maintenance planning, and further life-cycle cost analysis of bridges. In this work, a novel method is proposed for predicting the appropriate extreme traffic load distribution. Specifically, the average conditional exceedance rate (ACER) statistical model is estimated from the historical traffic loads which was collected through a weigh-in-motion system installed in toll stations. The basic idea of the ACER approach lies in the introduction of a cascade of conditioning approximations and the average exceedance rate to capture the dependence effects and obtain the data tail, the trend features of which are fitted with a similar Gumbel distribution function and extrapolated to the concerned level. An illustration case dealing with traffic loads using the ACER strategy is presented, the extreme value and confidence interval (CI) in any return period can be predicted by application of this approach. Furthermore, the peaks-over-threshold (POT) method based on the asymptotic extreme theory is also applied to illustrate the advantages of the ACER method. The ACER method has advantages in analyzing extreme traffic loads, with good robustness and the ability to handle extreme value prediction for different sampling strategies, it also can produce more accurate confidence intervals and predicts consistent extreme values. The study results are expected to help accurately determine traffic loads and ensure safety in bridge engineering.

Offshore wind speed assessment with statistical and attention-based neural network methods based on STL decomposition

This work proposes a novel offshore wind speed prediction approach by combining statistical method and attention-based neural network with seasonal-trend decomposition procedure with loess (STL). STL is utilized to decompose the processed data into season, trend and residual terms. Then, an attention-based long short-term memory neural network model (AT-LSTM), possessing the advantages of generalization and high-dimensional function approximation, is modeled to train season and residual terms with obvious volatility characteristics. And a hybrid model of auto-regressive integrated moving average (ARIMA) and LSTM is applied to predict the linear and nonlinear sequences in trend term with relatively gentle feature, yielding the proposed STL-AR-LSTM-ATLSTM model. Wherein, the proposed method is verified through sufficient pre-judgment experiments on season, trend and residual terms, as well as detailed multi-model comparison experiments. Finally, microcosmic prediction results and predicted statistical frequency distributions indicate that the new model has better prediction effect on offshore wind, compared to ARIMA, AT-LSTM, ARIMA-AT-LSTM models. Meanwhile, the presented model can reduce the lag problem of predicted values and perform well in the prediction of extreme values.

Bagged stepwise cluster analysis for probabilistic river flow prediction

River flow prediction plays a crucial role in flood control and water resource management. As an effective approach, the stepwise cluster analysis (SCA) has been well-accepted in river flow prediction but still encounters difficulties to acquire accurate predictions in sample-sparse regimes. This study proposes a new ensemble model by incorporating the bagging method with the SCA. The proposed Bagged Stepwise Cluster Analysis (BSCA) is applied to predict daily river flow in the Upper Min River, China. The results show that BSCA effectively improves the prediction accuracy of SCA. Particularly, it addresses the limited end clusters problem of SCA and thus enhances the extreme-value prediction. It also has a better overall prediction performance over the classical random forest and decision tree models. In addition, BSCA is capable of quantifying uncertainties through a probabilistic prediction, which can be used to assess risks of extreme flows and provide reliable river flow predictions. The proposed new model can be applied to other water systems and environmental engineering problems.

Station keeping of a subsea shuttle tanker system under extreme current during offloading

ABSTRACT This paper presents the station keeping challenge of the subsea shuttle tanker (SST) design during underwater loading and offloading at a subsea well under an extreme current environment. The paper investigates the movement of the SST during offloading with extreme current speeds, i.e. above 1.6 m/s, in the surge, heave and pitch motions, respectively. A linear quadratic regulator (LQR) is used for SST motion control. The LQR’s primary focus is to achieve the target for the SST during the offloading process. Then, the average exceedance rate method is used to predict the maximum and minimum potential depth excursion. This extreme value prediction result will serve as a basis for obtaining a cost-efficient design of the subsea shuttle tanker and provide recommendations for the decision-makers upon SST operation.

Extreme value prediction with modified Enhanced Monte Carlo method based on tail index correction

Extreme value prediction with modified Enhanced Monte Carlo method based on tail index correction

New extreme statistics strategy for the extreme value predictions of ship parametric rolling motions through limited model test observations

This work proposes a new extreme statistics strategy to predict the extreme parametric rolling angle through minimal observations from model tests. The core of this strategy is a new synthetic moment method. Based on the Hermite transformation and the Markov chain, the synthetic moment method can provide extreme value predictions for stationary processes. Besides, this work introduces one specific simplification to solve the problem induced by the non-stationarity of the parametric roll. With the help of the new extreme statistics strategy, this work obtains the CDF and PDF of the extreme parametric rolling angle for the C11 container ship. These predictions are derived from only one observed model test time history. The data statistics of other model tests and a vast amount of numerical simulations verify the validities of these predictions. Compared with other methods, the new strategy shows advantages for small sampling predictions. It is possible to replace the widely used ACER method in case of insufficient samples. In addition, further research confirms that the new strategy's intrinsic probability model differs from the widely used GEV model, which is more reasonable for extreme value predictions in a finite time range. The newly presented strategy should effectively predict the extreme value for other complicated stochastic responses.

Novel methods for reliability study of multi-dimensional non-linear dynamic systems

This research presents two unique techniques for engineering system reliability analysis of multi-dimensional non-linear dynamic structures. First, the structural reliability technique works best for multi-dimensional structural responses that have been either numerically simulated or measured over a long enough length to produce an ergodic time series. Second, a novel extreme value prediction method that can be used in various engineering applications is proposed. In contrast to those currently used in engineering reliability methodologies, the novel method is easy to use, and even a limited amount of data can still be used to obtain robust system failure estimates. As demonstrated in this work, proposed methods also provide accurate confidence bands for system failure levels in the case of real-life measured structural response. Additionally, traditional reliability approaches that deal with time series do not have the benefit of being able to handle a system's high dimensionality and cross-correlation across several dimensions readily. Container ship that experiences significant deck panel pressures and high roll angles when travelling in bad weather was selected as the example for this study. The main concern for ship transportation is the potential loss of cargo owing to violent movements. Simulating such a situation is difficult since waves and ship motions are non-stationary and complicatedly non-linear. Extreme movements greatly enhance the role of nonlinearities, activating effects of second and higher order. Furthermore, laboratory testing may also be called into doubt due to the scale and the choice of the sea state. Therefore, data collected from actual ships during difficult weather journeys offer a unique perspective on the statistics of ship movements. This work aims to benchmark state-of-the-art methods, making it possible to extract necessary information about the extreme response from available on-board measured time histories. Both suggested methods can be used in combination, making them attractive and ready to use for engineers. Methods proposed in this paper open up possibilities to predict simply yet efficiently system failure probability for non-linear multi-dimensional dynamic structure.

Feature Extraction and Prediction of Water Quality Based on Candlestick Theory and Deep Learning Methods

In environmental hydrodynamics, a research topic that has gained popularity is the transmission and diffusion of water pollutants. Various types of change processes in hydrological and water quality are directly related to meteorological changes. If these changing characteristics are classified effectively, this will be conducive to the application of deep learning theory in water pollution simulation. When periodically monitoring water quality, data were represented with a candlestick chart, and different classification features were displayed. The water quality data from the research area from 2012 to 2019 generated 24 classification results in line with the physics laws. Therefore, a deep learning water pollution prediction method was proposed to classify the changing process of pollution to improve the prediction accuracy of water quality, based on candlestick theory, visual geometry group, and gate recurrent unit (CT-VGG-GRU). In this method, after the periodic changes of water quality were represented by candlestick graphically, the features were extracted by the VGG network based on its advantages in graphic feature extraction. Then, this feature and other scenario parameters were fused as the input of the time series network model, and the pollutant concentration sequence at the predicted station constituted the output of the model. Finally, a hybrid model combining graphical and time series features was formed, and this model used continuous time series data from multiple stations on the Lijiang River watershed to train and validate the model. Experimental results indicated that, compared with other comparison models, such as the back propagation neural network (BPNN), support vector regression (SVR), GRU, and VGG-GRU, the proposed model had the highest prediction accuracy, especially for the prediction of extreme values. Additionally, the change trend of water pollution was closer to the real situation, which indicated that the process change information of water pollution could be fully extracted by the CT-VGG-GRU model based on candlestick theory. For the water quality indicators DO, CODMn, and NH3-N, the mean absolute errors (MAE) were 0.284, 0.113, and 0.014, the root mean square errors (RMSE) were 0.315, 0.122, and 0.016, and the symmetric mean absolute percentage errors (SMAPE) were 0.022, 0.108, and 0.127, respectively. The established CT-VGG-GRU model achieved superior computational performance. Using the proposed model, the classification information of the river pollution process could be obtained effectively and the time series information could also be retained, which made the application of the deep learning model to the transmission and diffusion process of river water pollution more explanatory. The proposed model can provide a new method for water quality prediction.

Improving extreme offshore wind speed prediction by using deconvolution

This study proposes an innovative method for predicting extreme values in offshore engineering. This includes and is not limited to environmental loads due to offshore wind and waves and related structural reliability issues. Traditional extreme value predictions are frequently constructed using certain statistical distribution functional classes. The proposed method differs from this as it does not assume any extrapolation-specific functional class and is based on the data set's intrinsic qualities. To demonstrate the method's effectiveness, two wind speed data sets were analysed and the forecast accuracy of the suggested technique has been compared to the Naess-Gaidai extrapolation method. The original batch of data consisted of simulated wind speeds. The second data related to wind speed was recorded at an offshore Norwegian meteorological station.

A Wave Directionality and a Within-Year Wave Climate Variability Effects on the Long-Term Extreme Significant Wave Heights Prediction in the Adriatic Sea

The extreme significant wave height predictions often neglect within-year wave climate variability and wave directionality. Depending on a geographical region, local wind patterns and year climate variability could have an influence on the long-term prediction of waves. The Adriatic Sea having two dominant wind patterns of different characteristics, Bura and Jugo, is a great example for the case study. The 23-year hindcast wave data used in the presented study is extracted from the WorldWaves database. Based on wind and wave data, annual extreme significant wave heights generated by different wind patterns and for different months are fitted by Gumbel distribution using maximum likelihood estimation. Combined long-term extremes are then predicted by calculating system probability. It was found that considering the wave directionality, and especially the seasonality of wave climate, leads to a larger prediction of extreme significant wave heights. The extreme value prediction considering wave directionality on average yields 4% larger significant wave heights, while considering within-year climate variability leads to, on average, 8% larger extremes compared to the predictions when both effects are neglected.

Cargo ship aft panel stresses prediction by deconvolution

This article introduces novel extreme value prediction method that can be used for a variety of offshore engineering applications. First, to demonstrate the novel method, fictitious data from a non-linear Duffing oscillator and measured wave heights were used as examples. The second incident included a container ship that experienced significant deck panel strains while traveling across the Atlantic Ocean in bad weather. The main concern for cargo ship transportation is potential loss of container owing to violent movements. It is challenging to model such a situation because waves and ship motions are both non-stationary and complicatedly nonlinear. Extreme motions greatly increase the role of nonlinearities, activating effects of second and higher order.Furthermore, due to the scaling and the choice of sea state, laboratory testing may also be called into doubt. Therefore, data collected from actual ships during difficult weather voyages offers a special perspective on the statistics of ship motions.This paper aims to highlight an alternative method of extrapolation that is based on intrinsic properties of the data set itself and does not assume any extrapolation functional class. Extreme value predictions typically originate from certain statistical distribution functional classes to fit the data and then extrapolate. Engineering design can make use of the unique extrapolation method that has been proposed. The proposed method's forecast accuracy has been verified in comparison to the Averaged Conditional Exceedance Rate (ACER) extrapolation method.

Novel deconvolution method for extreme FPSO vessel hawser tensions during offloading operations

Robust prediction of extreme hawser tensions during Floating Production Storage and Offloading (FPSO) operations is an important safety and reliability concern. Excessive hawser tension may occur during certain environmental conditions, posing an operational risk. The novel deconvolution method is proposed in this paper to provide an accurate extreme value prediction. The predicted return level values produced by the new deconvolution approach are compared with those obtained by the Naess-Gaidai method. On the basis of the suggested approach's overall performance, it is concluded that the innovative deconvolution method may give a more robust and accurate forecast of excessive hawser stress. The stated method may be used effectively at the stage of vessel design while determining appropriate vessel characteristics to reduce possible FPSO hawser stress.

Simplified models for uncertainty quantification of extreme events using Monte Carlo technique

The accurate prediction of extreme events based on measured data is an important task because it can facilitate infrastructure reliability design, risk assessment, and disaster mitigation. However, owing to limited sample size, considerable uncertainties are introduced to extreme value predictions with respect to different probabilities of exceedance. This paper proposes an uncertainty quantification analysis algorithm for extreme value statistics using the Monte Carlo technique. The study is focused on three widely used probability distributions fitted by the maximum likelihood method: generalized extreme value, generalized Pareto, and Pearson Type III. Three simplified models of the standard error for extreme value quantiles are explicitly developed as a function of the sample size, return period, and distribution parameters. These models eliminate the constraints resulting from the assumption of normality as well as provide more straightforward and accurate expressions that are convenient for engineering applications. The results indicate that the standard error increases with the return period as well as the shape and scale parameters of the distribution; however, it decreases with the sample size. The developed simplified models were applied to extreme event analyses of wind speed and precipitation at several typical sites. To clarify the practicability and rationality of the method, a wind hazard map for non-typhoon winds was developed for mainland China.

A feature-based adaptive combiner for coupling meta-modelling techniques to increase accuracy of river flow prediction

ABSTRACT Data-driven models are relatively simple but powerful methods for river flow prediction; to use the strengths of each, hybrid model-free methods are proposed. In this study, a feature-based adaptive combiner (FBAC) is introduced that uses features extracted from the data to provide a robust meta-model for river flow prediction. Artificial neural network (ANN), random forest regression (RFR), support vector regression (SVR), and a modified dynamic K-nearest neighbours (DKNN) are used to construct the FBAC. DKNN is incorporated to assess the flexibility of the FBAC in using novel data-driven approaches. The FBAC’s performance is evaluated using two years of the daily inflow to the Azad Reservoir in western Iran. Results indicate that the FBAC shows 31.8% and 29.5% improvement in accuracy, in terms of a reduction in the Root Mean Square Error (RMSE) value, compared to the best individual and combined model, respectively. For extreme value prediction, improvements are increased to 35.1% and 32.6%, respectively.

Prediction of extreme cargo ship panel stresses by using deconvolution

Extreme value predictions typically originate from certain functional classes of statistical distributions to fit the data and are subsequently extrapolated. This paper describes an alternative method for extrapolation that is based on the intrinsic properties of the data set itself and that does not pre-assume any extrapolation functional class. The proposed novel extrapolation method can be utilized in engineering design. To illustrate this, this study uses two examples to showcase the advantages of the proposed method. The first example used synthetic data from a non-linear Duffing oscillator to illustrate the new method. The second example was an actual container ship sailing between Europe and America and experiencing large deck panel stresses in severe weather. In this example, actual onboard measured data were used in the present study. This example represents a real and physical case that is challenging to model due to the non-stationary and highly non-linear natures of the wave-ship load responses. This is especially so in the case of extreme responses, where the roles of second and higher-order responses tend to be more prominent and have higher contributions. The prediction accuracy of the proposed method was also validated versus the Naess–Gaidai extrapolation method. Finally, this study discusses new methods for generic smoothing of distribution tail irregularities due to underlying scarcity in the data set.