Grid management is important for energy distribution, system security and market economics, and one of the key issues is accurate and stable prediction of wind speed for optimal operation and management of wind power connected to the grid. In this study, a novel two-layer hybrid method termed SSP-BO-LSTM is proposed for ultra-short-term wind speed prediction, such as four-hour ahead. The first layer is based on the smoothing spline preprocessing (SSP) method to remove non-Gaussian and non-stationary volatilities from the high-resolution wind speed series. Then, the processed wind speed data are predicted four-hour ahead by the long short-term memory (LSTM) model, and a bayesian optimization (BO) algorithm is presented to optimize the hyperparameters of the LSTM model. To evaluate the performance of the proposed SSP-BO-LSTM model, a case study of ultra-short-term wind speed prediction is conducted, including three high-resolution wind speed series from wind turbine measurements. Moreover, six other prediction models are introduced for in-depth comparison, and a comprehensive analysis is performed. The results show that the proposed model can improve the accuracy of four-hour ahead prediction by about 8%-35%, proving to be more effective and stable in providing acceptable results compared to the other six models mentioned in this study.

The flashover is one of the common incidents in transmission line systems. The wind-induced swing angle of the suspension insulator string is the key critical index for flashover incident limit state function. Based on the equivalent static wind load obtained from Gust Loading Envelope method, the wind-induced swing angle of suspension insulator string could be explicitly expressed by 10-min mean wind speed and line parameters, with consideration of wind speed correlation along conductor span. Therefore, the short-term forecast of the 10-min mean wind speed trend will be of great significance to the early warning of flashover incidents. This study proposes an improved hybrid prediction model based on the secondary data decomposition technology and neural network optimized by Bat algorithm for short-term multi-step-ahead wind speed prediction. In the improved hybrid prediction model, the high-frequency components of original wind speed data will be secondary decomposed because of greater prediction error. Then, the Bat algorithm is used further to optimize the initial weight and threshold parameters of the neural network to improve prediction accuracy. The accuracy and superiority of the proposed prediction model are verified by the application example. The obtained prediction results of wind speed will be substituted into the limit state function of flashover incidents to assess the flashover risk. The results show that the improved hybrid model has better performance in the multi-step-ahead forecast and can be used for flashover incident early warning.

Aeroelastic instability (i.e., flutter) is a critical issue that threatens the safety of flexible bridges with increasing span length. As a promising technique for flutter prevention, active aerodynamic control using auxiliary surfaces attached to the bridge deck (e.g., winglets and flaps) can be utilized to extract the stabilizing forces from the surrounding wind flow. Conventional controllers for the active aerodynamic control are usually designed using linear model-based schemes [e.g., linear quadratic regulator (LQR) and H-infinity control]. In addition to suffering from model inaccuracies, the obtained linear controller may not work well considering the high complexity of the inherently nonlinear wind-bridge-control system. To this end, this study proposes a nonlinear model-free controller based on deep reinforcement learning for active flutter control of long-span bridges. Specifically, a deep neural network (DNN), with the powerful ability to approximate nonlinear functions, is introduced to map from the system state (e.g., the motion of bridge deck) to the control command (e.g., reference position of the actively controlled surface). The DNN weights are obtained by interacting with the wind-bridge-control environment in a trial-and-error fashion (hence the explicit model of system dynamics is not required) using reinforcement learning algorithms of deep deterministic policy gradient (DDPG) due to its ability to tackle continuous actions with high training efficiency. As a proof of concept, numerical examples on active flutter control of a flat plate and a bridge deck are conducted to demonstrate the good performance of the proposed scheme.

Improving the accuracy of wind speed predictions is crucial to the scheduling plan and operating stability of the power grid system. However, few studies utilize the generative adversarial network (GAN) to implement wind speed predictions considering the influence of other meteorological factors. Additionally, the accuracy of wind speed predictions needs to be further improved, especially for multi-step wind speed predictions. Subsequently, a novel hybrid wind speed prediction model is proposed, including four modules: (1) data collection of the weather research and forecasting (WRF) simulation, (2) data generation of the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and GAN with the generator of bidirectional long short-term memory (BLSTM), (3) an error correction strategy of the CEEMDAN and GAN-BLSTM, and (4) hyperparameters optimization of the grid search (GS) and particle swarm optimization (PSO). Three datasets are utilized to validate the forecasting accuracy of the proposed model. The verification results demonstrate that the forecasting performance of the proposed model outperforms other baseline models. Taking the mean absolute percentage error (MAPE) of the ten-step prediction for the three datasets as an example, the MAPE values are respectively 0.51%, 0.46%, and 0.55% with correction, leading to 9.16%, 9.77%, 9.59% lower than those without correction. Above all, the proposed model possesses excellent wind speed prediction accuracy, especially in multi-step wind speed predictions, due to its lower values of MAPE with similar coefficients of determination (R²) values.

Wind speed measurement is one of the most fundamental tasks for multidiscipline applications and plays an important role in the design and maintenance of modern infrastructures. Wind speed is usually measured using conventional gauges which require additional connections to sensors or collection boxes, and their complex operating principles make these devices largely serve only professionals. This study proposed a novel framework associated with a machine learning architecture to estimate wind speed directly from acoustic signal collected using smartphones. The one-dimensional convolutional network is employed to characterize the underlying relationship between the frequency domain features of the acoustic signal and wind speed. An experimental dataset is collected in wind tunnel laboratory in which the wind speed is measured using cobra probe and the acoustic signal is recorded using smartphone. The influence of encountering direction angle on the 1D-CNN wind speed measurement model is also discussed, as well as the ability of the model to resist noise. The favorable robustness and generalization performance of the 1D-CNN model are verified from multiple perspectives, illustrating the feasibility and practical value of using smartphones to measure wind speed.