The super-turbine wind power conversion paradox: using machine learning to reduce errors caused by Jensen's inequality

Abstract

Abstract. Wind power is a variable generation resource and therefore requires accurate forecasts to enable integration into the electric grid. Generally, the wind speed is forecast for a wind plant and the forecasted wind speed is converted to power to provide an estimate of the expected generating capacity of the plant. The average wind speed forecast for the plant is a function of the underlying meteorological phenomena being predicted; however, the wind speed for each turbine at the farm is also a function of the local terrain and the array orientation. Conversion algorithms that assume an average wind speed for the plant, i.e., the super-turbine power conversion, assume that the effects of the local terrain and array orientation are insignificant in producing variability in the wind speeds across the turbines at the farm. Here, we quantify the differences in converting wind speed to power at the turbine level compared with a super-turbine power conversion for a hypothetical wind farm of 100 2 MW turbines as well as from empirical data. The simulations with simulated turbines show a maximum difference of approximately 3 % at 11 m s−1 with a 1 m s−1 standard deviation of wind speeds and 8 % at 11 m s−1 with a 2 m s−1 standard deviation of wind speeds as a consequence of Jensen's inequality. The empirical analysis shows similar results with mean differences between converted wind speed to power and measured power of approximately 68 kW per 2 MW turbine. However, using a random forest machine learning method to convert to power reduces the error in the wind speed to power conversion when given the predictors that quantify the differences due to Jensen's inequality. These significant differences can lead to wind power forecasters overestimating the wind generation when utilizing a super-turbine power conversion for high wind speeds, and indicate that power conversion is more accurately done at the turbine level if no other compensatory mechanism is used to account for Jensen's inequality.

Examining the Potential of a Random Forest Derived Cloud Mask from GOES-R Satellites to Improve Solar Irradiance Forecasting

“…The final prediction for a continuous variable, such as cloud fraction, is the mean of the instances in the final leaf node for an instance that follows the rules of the branches down to the final leaf. This is illustrated by the green decision nodes in Figure 2 that depicts a random forecast model, which is described in further detail in [23]. In this illustration, the darker boxes indicate how a RF model would make a prediction for a given instance by following the set of rules in each tree and computing the ensemble average of the prediction from each tree in the forest.…”

A database of hourly wind speed and modeled generation for US wind plants based on three meteorological models

“…In multi-turbine wind plants, the use of a power curve does not account for losses that arise due to turbine wakes. Also, because we limit our wind speed estimates to a single time series for each plant and meteorological model, our estimate does not account for variation of wind speeds across turbines within each plant, which is important to consider given the non-linear nature of turbine power curves 51 . Further, some manufactures have power curves with characteristics that are slightly different from our basic description, such as mildly downward sloping curves above the rated speed.…”

Analysis of machine repair time prediction using machine learning at one of leading footwear manufacturers in Indonesia

“…Several supervised machine learning classification method algorithms were chosen to get the best accuracy value [9], [10]. The algorithms are logistic regression (LR) [11], [12], naive bayes (NB) [13], [14], random forest (RF) [15], [16], k-nearest neighbor (KNN) [17], [18], and support vector machine (SVM) [19], [20]. Selected five classifications supervised machine learning based on each algorithm have on different dimension metrics there are parametric-simple for LR and NB, parametric-complex for SVM, non-parametric-simple for KNN, and non-parametic-complex for RF.…”