Models for monitoring wind farm power

Abstract

Different models for monitoring wind farm power output are considered. Data mining and evolutionary computation are integrated for building the models for prediction and monitoring. Different models using wind speed as input to predict the total power output of a wind farm are compared and analyzed. The k-nearest neighbor model, combined with the principal component analysis approach, outperforms other models studied in this research. However, this model performs poorly when the conditions of the wind farm are abnormal. The latter implies that the original data contains many noisy points that need to be filtered. An evolutionary computation algorithm is used to build a nonlinear parametric model to monitor the wind farm performance. This model filters the outliers according to the residual approach and control charts. The k-nearest neighbor model produces good performance for the wind farm operating in normal conditions.

Introduction

The generation of wind energy on an industrial scale is relatively new. It is then natural that the performance of wind power farms has not been adequately studied. One of the weakest points in wind power generation is the low predictive accuracy of the energy output. Like industrial corporations managed by enterprise-wide systems, a software solution for prediction of wind farm performance (including the amount of energy produced) is needed. The envisioned wind farm performance prediction models should be able to predict the amount of energy produced on different time scales, e.g., 10 min, 1 h, a day, etc. Such models could transform a wind farm into a wind power plant.

Researchers have applied different methodologies in studying wind farms. Cameron and Michael [3] combined the fuzzy set and neural network approaches in an adaptive-neurons-fuzzy inference system to forecast a wind time series. Landberg [6] builts a model to predict the power produced by a wind farm using the data from the weather prediction model (HIRLAM) and the local weather model (WASP). Li et al. [13] compared regression and neural network (NN) models in order to estimate a turbine's power curve. They reported that the NN model outperformed the regression model. Goh et al. [11] proposed a neural network architecture, the complex-valued pipelined recurrent neural network (CPRNN) using a complex value (combined wind speed and direction into one complex value) as input, for predicting the turbine output. Santoso and Le [14] focused on modeling fixed-speed wind turbines. They modeled the component blocks of a turbine (for aerodynamic, mechanical, and electrical components), and aggregated them into models of a single turbine and a wind farm. Lange and Focken [28] presented various models for short-term wind power prediction, including physics-based, fuzzy, and neuro-fuzzy models. Barbounis et al. [29] constructed a local recurrent neural network model for long-term wind speed and power forecasting based on the meteorological data. Hourly forecasts for up to 72 h ahead were produced for a wind park.

Wind energy has become one of the most important sources of energy. Building accurate models for predicting power output and health monitoring of wind farms is needed by this new industry. Developing such models is challenging, as a large number of parameters are involved. The high dimensional and stochastic nature of a wind farm environment calls for new modeling approaches. The developments in data mining (DM) and evolutionary computation (EC) offer promising approaches to model wind farms. Numerous applications of data mining in manufacturing, marketing, medical informatics and the energy industry have proven to be effective in support of decision making [2], [5], [7], [8], [24]. Successful applications of evolutionary computation have also been reported in many other domains [1], [4], [9], [10], [12], [15].

In this paper, a variety of different approaches, including data mining, evolutionary computation, principal component analysis (PCA), residual approach, and control charts, have been used to build prediction models and characterize power curves of a wind farm by a nonlinear parametric model. The models are built using historical data collected by SCADA (Supervisory Control and Data Acquisition) systems at a wind farm.