Search

Flutter may lead to collapse of long-span bridges and need to be avoided, which makes the estimation of the onset flutter essential for long-span bridge design. However, the efficiency of traditional solution methods of predicting onset flutter is not high. In this study, it is converted into an optimization model, which is solved by a proposed advanced particle swarm optimization (APSO) with a novel inertia weight strategy and mutation mechanism. Two well-known benchmark functions are firstly employed to validate the performance of APSO. Compared with other existing PSO methods, APSO not only has high accuracy, but also has the fastest convergence speed. Then, APSO is used to find the flutter critical wind speed of bridges. A good agreement is obtained comparing predictions on the onset flutter by the proposed methods and available tests on several study cases. In addition, the value range of APSO variables is discussed in depth. It is confirmed that the proposed APSO algorithm has excellent stability, robustness and fast convergence, which is suitable for searching onset flutter of bridges. [ABSTRACT FROM AUTHOR]

Precise forecasting of wind speed is vital to guarantee safe travel on bridges. Nevertheless, current research primarily concentrates on single-point prediction. When it comes to forecasting spatial wind fields, graph neural networks prove to be powerful methods, but many of them have yet to take into account the impact of frequency domain attributes. To address this, a novel model called the graph neural network with time-varying frequency features (GTF) is proposed. Its core layer employs wavelet transforms in matrix form to extract various frequency sub-series, followed by the use of adaptive adjacency matrices to identify their spatial characteristics. Furthermore, aggregating all frequency band information and exploring spatial features further through multi-kernel convolutions. The predictive performance of different models for wind speed forecasting is assessed by utilizing spatial wind field data collected on a large-span bridge. Experimental results demonstrate that spatiotemporal model GTF outperforms the other models in terms of predictive accuracy. The quantity of wind observation points and their level of correlation exert a certain impact on the GTF model, while hyperparameters such as stacking depth and wavelet decomposition parameters have minimal effects. Finally, recommended parameters for the GTF model are provided. • A graph neural network involving time-varying frequency domain features is proposed. • Atual spatial wind field measurements are utilized for evaluating the proposed model. • The impact of spatial wind field characteristics and model parameters were examined. • The suggested model exhibits commendable precision and stability. [ABSTRACT FROM AUTHOR]

Turbulence coherence in mountainous areas necessitates meticulous evaluation due to the complex terrain blocking. This study focuses on the turbulence coherence of the non-uniform wind field in a canyon via the anemometers positioned along a bridge span. Based on a proposed multi-point clustering, two categories of strong winds impacted by a nearby ridge are identified. The turbulence coherence of the two strong winds is then investigated with an extra focus on the phase spectrum. Several complex coherence functions are extended from commonly employed real forms, including the simplified Krenk's formula proposed. The fitting results of them all suggest that parameters such as the decay factor of the two strong winds show a negative linear correlation with separation but their relationship with wind speed is not significant. For the phase spectrum, in addition to skew winds, the contribution of horizontal wind shear is first emphasized in the two strong winds with piecewise fittings. This paper also discusses single-point coherence between the along-wind and vertical turbulence components. It is found to be unusually positive in sign due to the influence of the local topography and a formula with a constant term is suggested that substantially improves the fitting compared to Solari's formula. • The turbulence coherence with different separations is studied in a non-uniform wind field. • The strong winds are classified by multi-point clustering of the orthogonal components. • The contributions of horizontal wind shear and skew wind to the phase spectrum are emphasized. • The single-point coherence between different turbulence components is investigated. [ABSTRACT FROM AUTHOR]

Wind speed prediction is a crucial element of effective wind energy utilization, necessitating the use of probabilistic wind speed prediction to facilitate practical decision-making. However, most current deep learning generative models neglect to incorporate frequency information. To address this issue, this study proposes a generative model neural network that utilizes multi-feature fusion and Gaussian mixture techniques. The proposed model leverages wavelet transformation, which is represented as a weight matrix, in combination with several convolution operations to extensively explore the fundamental features of wind speed time series. Prior to multi-step prediction, target values are modeled as a multi-dimensional Gaussian mixture distribution, with the extracted characteristics mapped onto crucial distribution parameters. Monte Carlo sampling is then used to complete the wind speed probability prediction. Furthermore, the study develops a loss function and a two-stage training strategy for this network. The performance of the proposed model is compared to that of ten state-of-the-art models using actual wind speed time series and six typical error indices. The results demonstrate that the proposed model provides both more accurate deterministic predictions and more meaningful probabilistic forecasts than other examined models for various prediction. • A generative neural network with multi features fusion and Gaussian mixture is built. • It presents a flexible matrix formulation of the wavelet transform, along with a derived loss function and training method. • The deterministic and probabilistic prediction accuracy of the given model has been verified. • The stability of the model remains robust regardless of the selection of wavelet function or Gaussian mixture components. [ABSTRACT FROM AUTHOR]

• The upper and lower surfaces of the deck are the vibration-induced parts. • The phase raise of lift leads to unsteady galloping instability. • Barriers' porosity or height change aerostability while their processes differ. • Mechanical damping is not a decisive parameter affecting critical wind speed. The long-span bridges with truss beam are widely recognized as typical wind-sensitive structures, and its wind-induced divergence events may become a very predominant issue. This work evaluates the galloping performance of a typical truss beam with wind barriers, which is detected to be susceptible to the infrequent vertical instability at specific attack angles, as evidenced by experimentation and numerical analysis. Since the truss beam constitutes a complex cross section, this study employs energy analysis to distinguish the crucial parts affected by vibration and discovers that the galloping vibration is mostly caused by the surfaces of the deck. Rather than relying on conventional quasi-steady theory, this study explores the unsteady aerodynamic forces to identify the underlying galloping driving mechanism. Our findings reveal that the instigator of galloping instability is the phase rising of lift with increasing wind speed. Additionally, the nonlinear mechanical damping is not a decisive parameter. Furthermore, we have investigated the impact of an essential ancillary facility, the wind barrier, and discovered that its galloping stability can be enhanced by increasing its porousness and decreasing its height. However, the underlying mechanisms behind this improvement are distinct. [ABSTRACT FROM AUTHOR]

A growing number of long-span bridges are under construction across straits or through valleys, where the wind characteristics are complex and inhomogeneous. The simulation of inhomogeneous random wind velocity fields on such long-span bridges with the spectral representation method will require significant computation resources due to the time-consuming issues associated with the Cholesky decomposition of the power spectrum density matrixes. In order to improve the efficiency of the decomposition, a novel and efficient formulation of the Cholesky decomposition, called "Band-Limited Cholesky decomposition," is proposed and corresponding simulation schemes are suggested. The key idea is to convert the coherence matrixes into band matrixes whose decomposition requires less computational cost and storage. Subsequently, each decomposed coherence matrix is also a band matrix with high sparsity. As the zero-valued elements have no contribution to the simulation calculation, the proposed method is further expedited by limiting the calculation to the non-zero elements only. The proposed methods are data-driven ones, which can be applicable broadly for simulating many complicated large-scale random wind velocity fields, especially for the inhomogeneous ones. Through the data-driven strategies presented in the study, a numerical example involving inhomogeneous random wind velocity field simulation on a long-span bridge is performed. Compared to the traditional spectral representation method, the simulation results are with high accuracy and the entire simulation procedure is about 2.5 times faster by the proposed method for the simulation of one hundred wind velocity processes. [ABSTRACT FROM AUTHOR]

Highlights • A novel prediction framework is proposed. • Three new hybrid models based on the framework are put forward. • Compared to normal methods, the proposed models yield a better prediction accuracy. Abstract In this paper, a novel framework for wind speed forecasting is proposed. In the new prediction framework, wavelet transform is firstly adopted to decompose original wind speed history into several sub-series. Then, for low-frequency sub-series, recurrent neural networks are used to extract deeper features, which are fed into suitable machine learning methods for predicting, while others are still predicted by normal methods. Meanwhile, three new hybrid models are established, where support vector machine is taken as the predictor, and the standard recurrent neural network and its variant version: long short term memory neural networks and gated recurrent unit neural networks are employed to extract the deeper features. Four experiments from the real world are conducted through the proposed methods as well as normal algorithms. The results demonstrate that the three new proposed hybrid models based on the novel framework yield more accurate predictions. [ABSTRACT FROM AUTHOR]

To raise the wind speed prediction accuracy, Wavelet Transform (WT) is widely employed to disaggregate an original wind speed series into several sub series before forecasting. However, the highest frequency sub series usually has a great disturbance on the final prediction. In the study, for raising the forecasting accuracy, Singular Spectrum Analysis (SSA) is applied to make further processing on the highest frequency sub series, instead of making no modification on or getting rid of it. So a hybrid decomposition technology called Improved WT (IWT) is proposed. Meanwhile, a new hybrid model IWT-ENN combined with IWT and Elman Neural Network (ENN) is also designed. The procedure of IWT is systematically investigated. Experimental results show that: (1) the performance of the hybrid model IWT-ENN has a great improvement compared to that of others including the persistence method, ENN, Auto-Regressive (AR) model, Back Propagation Neural Network (BPNN) and Empirical Mode decomposition (EMD)-ENN; (2) compared to the two general strategies where the highest frequency sub series is without retreatment or eliminated, the new proposed hybrid model IWT-ENN has the best prediction performance. [ABSTRACT FROM AUTHOR]

For the sensible and efficient use of wind energy, accurate wind speed forecast is crucial. To improve the accuracy of short-term wind speed prediction, a novel recurrent neural network known as the time-frequency recurrent neural network, or TFR for short, is developed. The wavelet transformation is naturally incorporated into the TFR architecture in order to mine the time-frequency characteristics. Additionally, the convolution processes are combined to extract the inherent correlation of time series, enhancing the TFR's performance and creating an advanced model known as CNN-TFR. The prediction ability, parameter sensitivity, and training time of the suggested models for multi-step wind speed forecasts are examined using the wealth of wind speed data from a genuine observation site. It is found that TFR offers greater prediction performance as compared to conventional recurrent neural networks since it can access frequency domain knowledge. Additionally, CNN-TFR's prediction performance has been further improved, making it superior to other CNN based models. For the proposed CNN-TFR model, its sensitivity to input length and wavelet parameters is investigated. It has been shown that with little training time, the CNN-TFR model with strong and robust prediction ability can be utilized to anticipate real wind speed. • A novel recurrent neural network named TFR with mining frequency features is built. • The convolutional processes and TFR are merged to develop a new model. • The prediction accuracy and robustness of models are carefully checked and verified. [ABSTRACT FROM AUTHOR]

• End-to-end models with wavelet transformation and neural networks are created. • A two-stage training optimization approach is presented. • The proposed models are validated via real wind speed dataset. • The decomposition parameter of db4 and 1level is advised for the suggested models. Accurate wind speed forecasts are essential for wind energy applications. It is challenging to accurately anticipate wind speed because of its high volatility and non-stationarity. Although wavelet transformation-based hybrid models, the most representative hybrid models, are extensively used, the endpoint effect and fixed decomposition parameters limit their prediction accuracy. To enhance the performance of such models, a special end-to-end model architecture combined with wavelet decomposition in matrix form and recurrent neural networks (RNNs) is proposed. In addition, a two-stage training optimization approach is presented to prevent the characteristics of the wavelet decomposition coefficients from vanishing during model training. In this framework, three conventional RNNs, consisting of the standard RNN, the long short-term memory network (LSTM), and the gated recurrent units network (GRU), are utilized to construct three innovative models, referred to as MW-RNN, MW-RNN, and MW-GRU, respectively. Then, using two years of data from a site, the recommended models are contrasted with the standard ones. The results show that the recommended models beat traditional ones in a variety of error indices with just a slight increase in training costs. Additionally, the impacts of wavelet decomposition parameters on forecast accuracy are contrasted, and suggested settings for actual wind speed predictions are offered. [ABSTRACT FROM AUTHOR]

The joint probability density function (JPDF) of the wind field parameters is essential for wind load analysis or energy assessment, but one key parameter of the wind attack angle is often excluded in current studies. Three different methodologies are proposed, aiming to estimate the JPDF of wind speed, wind direction and wind attack angle. Strategy I consists of the vine approach with the modified binary Bernstein copula adopted for nonparametric estimation. The modified ternary Bernstein copula is recommended to estimate the ternary copula directly in Strategy II. In strategy III, the Offset Elliptical Normal mixture model is extended to evaluate the three-dimensional case, where the wind vector is resolved into three orthogonal components. A typical mountainous bridge site with strongly inhomogeneous wind characteristics is selected to inspect the performance of these strategies. Satisfactory results are given by all these strategies in which the strategy II has better goodness-of-fit. Strategy I is slightly less reliable in estimating the dependencies of higher tree variables while avoids the time-consuming nonparametric estimation of three-dimensional copula. The strategy III is based on the physical model and performs better in the conditional probability estimation when the number of samples is small. • The joint distribution of key parameters of wind characteristics is estimated. • Three general modeling strategies are proposed and compared. • The nonparametric estimation of ternary copula is realized by Bernstein copula. • The Offset Elliptical Normal mixture model is extended to 3-dimensional case. [ABSTRACT FROM AUTHOR]

Accurately predicting wind speed is crucial towards maximizing the use of wind energy. Unfortunately, due to the non-stationary and intermittent nature of wind speed, reliable prediction has proven to be challenging. Properly exploring the inherent sequence connection between wind speed and its underlying frequency characteristics would be advantageous in enhancing forecast accuracy. Expanding on this concept, innovative models have been developed. Initially, they utilize multiple convolutions to capture the internal features of the time series data. This is followed by the application of the state frequency memory recurrent block to extract the relevant frequency properties. To facilitate the identification of crucial hidden elements, different attention mechanisms are utilized between these two modules which enables multi-resolution feature fusion. Real-world data is used to conduct multi-step short-term wind speed prediction tests, and various classical models are evaluated. The results indicate that our proposed models, which possess time-frequency mining capability, outperform other models in terms of forecast accuracy across four commonly used error indices. Additionally, a parameter sensitivity analysis is conducted on the prediction models to demonstrate the high stability of the proposed model. Furthermore, the working mechanism of our proposed model is thoroughly explored in order to confirm its exceptional prediction accuracy. [ABSTRACT FROM AUTHOR]