Condition monitoring and performance forecasting of wind turbines based on denoising autoencoder and novel convolutional neural networks

Energy Reports, Journal Year: 2021, Volume and Issue: 7, P. 6354 - 6365

Published: Oct. 4, 2021

With the proportion of wind power in grid increasing, monitoring and maintenance turbines is becoming more important for reliability grid. In this study, a data-driven modelling framework based on deep convolutional neural networks constructed condition (CM) performance forecasting (PF). For CM, robust denoising autoencoder (DAE) model introduced to output reconstruction error (RE) raw signals. The RE processed state indicator by exponentially weighted moving average (EWMA) monitored control chart. PF, two multi-steps ahead models are generator bearing transformer temperature. To prevent overfitting caused abundant features, marginal effect analysis random forests implemented measure importance features. Besides, novel residual attention module (RAM) training strategies used improve their representation DAE PF models. Experiments real turbine dataset prove effectiveness proposed methods.

Language: Английский

---

Comprehensive comparison of various machine learning algorithms for short-term ozone concentration prediction

Alexandria Engineering Journal, Journal Year: 2021, Volume and Issue: 61(6), P. 4607 - 4622

Published: Oct. 22, 2021

Ozone (O3) is one of the common air pollutants. An increase in ozone concentration can adversely affect public health and environment such as vegetation crops. Therefore, atmospheric quality monitoring systems were found to monitor predict concentration. Due complex formation influenced by precursors meteorological conditions, there a need examine evaluate various machine learning (ML) models for prediction. This study aims utilize ML including Linear Regression (LR), Tree (TR), Support Vector (SVR), Ensemble (ER), Gaussian Process (GPR) Artificial Neural Networks Models (ANN) tropospheric using dataset. The dataset was created observing hourly average data from 3 different stations Putrajaya, Kelang, KL sites Peninsular Malaysia. prediction have been trained on this validated optimizing their hyperparameters. Additionally, performance evaluated terms RMSE, MAE, R2, training time. results indicated that LR, SVR, GPR ANN able give highest R2 (83 % 89 %) with specific hyperparameters Kelang KL, respectively. On other hand, SVR ER outweigh (79 Putrajaya station. Overall, regardless slightly differences, several developed learn patterns well provide good RMSE MAE. regression balance between high accuracy low time thus considered feasible solution application scenario.

Language: Английский

---

Wind power prediction based on PSO-Kalman

Energy Reports, Journal Year: 2022, Volume and Issue: 8, P. 958 - 968

Published: Feb. 18, 2022

Because of its clean and green, wind power is broadly used all over the world. Wind random unstable, so integration will inevitably bring great impact to system. Accurate prediction can effectively alleviate caused by uncertainty. In order increase accuracy prediction, this article uses paper swarm optimization algorithm (PSO) improve traditional Kalman filter, PSO-Kalman point model established. The proposed solves problem low filter observation noise process noise. Finally, based on error, non-parametric kernel density estimation for interval prediction. By experimental simulation, comparing error evaluation indexes it be found that smallest, indicating PSO Kalman. On basis, performance also better than before. Moreover, in converges fast has general applicability.

Language: Английский

---

Ensemble forecaster based on the combination of time-frequency analysis and machine learning strategies for very short-term wind speed prediction

Electric Power Systems Research, Journal Year: 2022, Volume and Issue: 214, P. 108863 - 108863

Published: Oct. 8, 2022

Language: Английский

---

Multi-step ahead wind speed prediction based on a two-step decomposition technique and prediction model parameter optimization

Energy Reports, Journal Year: 2022, Volume and Issue: 8, P. 6086 - 6100

Published: May 7, 2022

Accurate and reliable wind speed prediction is essential for the exploitation utilization of energy. In this paper, a novel hybrid multi-step model proposed based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), robust local mean (RLMD), improved whale optimization algorithm (IWOA), long-term short-term memory network (LSTM). CEEMDAN utilized to decompose series into number intrinsic functions, RLMD used second step most non-stationary function product functions. After two-step decomposition, group new subsequences formed. The (LSTM) constructed every subsequence an (IWOA) optimize key parameters affecting performance LSTM model. And at last, results are superimposed provide final results. effectiveness advancement verified by employing data from two different farms. according experimental comparison, it can be found that better than seven compared models.

Language: Английский

---