Monitoring WTGs using SCADA data




Wind Turbines Performance Evaluation Strategies On The Basis Of Raw SCADA Data

At Ventus we have developed four methods to evaluate the performance of WTG over time using Power, Wind Speed and Ambient Temperature SCADA measurement. Identifying a change in the performance of a Wind Turbine Generator (WTG) using the Raw SCADA Data may not be a simple task, particularly because the variability of the 10-minutes values during normal operation is quite wide. We named these methods ‘Power Residuals’, ‘Health Value -PC2 Dev’, ‘Quantiles’ and ‘Power Curves Evolution’ and in each we calculate a Key Performance Indicator (KPI). These KPI’s can be useful to identify changes or trends in the operation of the turbines, assess an improvement in the performance of the WTG after maintenance is done and help in the detection and prevention of possible failures in components which are directly related to the performance of the turbines. Read more (sorry no link for this one)

  1. Introduction

At Ventus we have developed four methods to evaluate the performance of WTG over time using Power, Wind Speed and Ambient Temperature SCADA measurement. Identifying a change in the performance of a Wind Turbine Generator (WTG) using the Raw SCADA Data may not be a simple task, particularly because the variability of the 10-minutes values during normal operation is quite wide. We named these methods ‘Power Residuals’, ‘Health Value -PC2 Dev’, ‘Quantiles’ and ‘Power Curves Evolution’ and in each we calculate a Key Performance Indicator (KPI). These KPI’s can be useful to identify changes or trends in the operation of the turbines, assess an improvement in the performance of the WTG after maintenance is done and help in the detection and prevention of possible failures in components which are directly related to the performance of the turbines, e.g. anemometers. An algorithm to automatically identify the changes in the KPIs is used to detect the statistical changes in the mean values of the KPI’s.

  1. SCADA Data

For the calculation of the KPI’s we use three 10-minutes signals of the SCADA of the WTG: Average Power, Average Wind Speed and Average Ambient Temperature. A filtering is done to obtain the data during the ‘normal’ operation of the WTG. Still after filtering, there is a wide dispersion in the Power vs Wind Speed measurements.

  1. Power Residuals

This method is based on the comparison between the Power measurements of the WTG and the Power obtained from a model of the operation of the WTG. To create the model a Gaussian Process Algorithm is applied to the data of a training period, to ‘learn’ the relationship between Power (output) and Wind Speed and Ambient Temperature (inputs). We then apply this model to the 10-minutes values of the rest of the period of operation, and calculate the difference between the Power obtained from the model and the actual Power. The ‘Power Residuals’ are then monitored over time, looking for deviations in that difference. In order to mitigate the variability of the Residuals we calculate a simple moving average and exponentially weighted moving average (EWMA), and monitor these instead.

  1. Health Value -PC2 Dev

We calculate and analyse this KPI as follows: Given the series of Power and Wind Speed measured in the nacelle, the method consists in evaluating in windows of time of 1 week the evolution of the second eigenvalue (d2) associated to these data, contemplating the period formed by a base period of three weeks plus the week of analysis. The first eigenvalue (d1) represents the variability along the main direction of scatter of Power-Velocity, while the second eigenvalue shows variations in the direction transversal to the scatter.

  1. Quantiles

The aim of this KPI is to evaluate over time the Power production of the WTG regardless of the measured wind speed value at each 10-minute period. The method consists in filtering the Power data for each occurring wind speed, binned within a range with values rounded to decimals, e.g. 0.1m/s. Then in each bin the Power values are sorted from higher to lower, assigning a quantile value (between 0 and 1) to each 10-minute period according to its position in this series. Due to high variability of the quantiles, we calculate a moving average and then monitor, looking for changes or tendencies over time. Figure 3. KPI: Quantiles for two wind turbines of the same wind farm.

  1. Power Curves Evolution

This algorithm consists in adjusting the data Power-Speed in windows of time of one week to a curve and then to compare the curves obtained in each week. We perform the adjustment of the data to a curve in a defined speed range, using the Penalized Spline Regression method.

  1. Change detector algorithm

To monitor each KPI, a change detection algorithm is used. The algorithm intends to detect changes in statistical indicators of data series over time, such as the mean value, while the maximum number of changes (MNC) to detect is an input.