Wind Power Deterministic Prediction and Uncertainty Quantification Based on Interval Estimation

Abstract

With the increasing penetration of wind power into modern power systems, accurate forecast models play a crucial role in large-scale wind power consumption and power system stability. To improve the accuracy and reliability of ultrashort-term wind power prediction, a novel deterministic prediction model and uncertainty quantification with interval estimation were proposed in this study. In consideration of the dynamic characteristics of a generator and conditional dependence, the generator rotor speed and pitch angle were regarded as the indicators of the dynamic characteristics of the generator, and light gradient boosting machine (LGBM) with a Bayesian optimization method was explored to build the deterministic prediction model. Considering the conditional dependence between output power and forecast error, A fuzzy C-means clustering method was used to cluster forecast errors into different clusters, and the best error probability distribution was obtained by fitting the error histogram with nonparametric kernel density estimation. Prediction intervals at different confidence levels were calculated and the error certainty was quantified. A case study was conducted to compare prediction accuracy and reliability by using the present and proposed methods. Results demonstrate that the LGBM deterministic prediction model combined with Bayesian optimization has better prediction accuracy and lower computational cost than the comparative models, specifically when the input features are high-dimensional big data. The nonparametric estimation method with conditional dependence is reliable for interval prediction. The proposed method has a certain reference value for wind turbines participating in frequency regulation and power control of power grid.

An Adaptive, Data-Driven Stacking Ensemble Learning Framework for the Short-Term Forecasting of Renewable Energy Generation

“…A surrogate function is built to express assumptions about the function to be optimized, and an acquisition function is selected to determine the next evaluation point. In this paper, the tree Parzen estimator (TPE) is employed to model the densities using a kernel density estimator, instead of directly modeling the objective function F by a probabilistic model p( f |D ) [47,48]. More details about Bayesian optimization are discussed in [45][46][47][48].…”

“…In this paper, the tree Parzen estimator (TPE) is employed to model the densities using a kernel density estimator, instead of directly modeling the objective function F by a probabilistic model p( f |D ) [47,48]. More details about Bayesian optimization are discussed in [45][46][47][48].…”

“…16, x FOR PEER REVIEW 8 of 20 density estimator, instead of directly modeling the objective function F by a probabilistic model ( ) p f D[47,48]. More details about Bayesian optimization are discussed in[45- 48].…”

A Two-Stage Multistep-Ahead Electricity Load Forecasting Scheme Based on LightGBM and Attention-BiLSTM

“…In addition, LightGBM can prevent overfitting by adjusting colsample by tree and subsample hyperparameters. LightGBM has been used for various time series forecasting tasks, such as electricity load forecasting [41,42] and wind power forecasting [43], and its single-output forecasting has been proven to be fast and accurate. As we need a fast and accurate single-output forecasting model in the first stage, we construct it using LightGBM.…”