Stefano Grassi

Statistical Learning Approach for Wind Speed Distribution Mapping: The UK as a Case Study

Wind resource assessment is fundamental when selecting a site for wind energy projects. Wind speed is influenced by a plethora of environmental factors and understanding its spatial variability is key for determining the economic viability of a site. Deterministic estimation methods, which are based on physics, represent the industry standard in wind-speed mapping. Over the years, these methods have proven capable of estimating wind speed with a relatively high accuracy. However measuring stations, which provide the starting data for all wind speed estimations, are often located at a distance from each other, in some cases, tens of kilometres or more. This adds an unavoidable level of uncertainty to the estimates, which deterministic methods fail to take into account. For this reason, even though there are ways of determining the overall uncertainty of the estimation, e.g. cross-validation, deterministic methods do not provide means of assessing the site-specific uncertainty. This paper introduces a statistical method for estimating wind speed, based on spatial statistics. In particular, we present a statistical learning approach, based on ensembles of regression trees, to estimate both the wind distribution in specific locations and to assess the site-specific uncertainty.

Optimal placement of wind turbines on a continuous domain: An MILP-based approach

Optimizing the placement of wind turbines in a defined region is a challenging task as it directly impacts the long term electricity production and thus the financial income of the investors. The main drivers affecting the cash flow are long-term wind resource, wind turbines location, investment upfront costs, O&M and financial framework (taxes, subsidies, depreciation, etc.). In this paper an iterative algorithm based on Mixed Integer Linear Programs (MILPs) is presented that allows to identify the wind turbines position that maximizes the long term cash flow. The Jensens wake effect model is used to describe the interaction between wind turbines. Furthermore, Geographic Information System (GIS) software and data are used to give a more realistic representation of the area where wind turbines can be installed. In order to validate the MILP-based approach, a wind farm in Kansas is used as case study and the actual internal rate of return (IRR) resulting from the existing wind turbines layout is compared to the one obtained using the proposed methodology.

A GIS-Based Process for Calculating Visibility Impact from Buildings During Transmission Line Routing

Planning linear infrastructures can be a tedious task for regions characterized by complex topography, natural constraints, high density population areas, and strong local opposition. These aspects make the planning of new transmission lines complex and time consuming. The method proposed in this work uses Multi-Criteria Analysis and Least-Cost Path approaches combined with a viewshed analysis in order to identify suitable routes. The visual impact is integrated, as a cost surface, into the process and combined with natural and anthropological constraints. The cumulated visibility of each raster cell is estimated as the sum of the weighted distance between buildings and the cell itself. In order to reduce the typical zig-zags resulting from Least-Cost Path methods, a weighted straightening approach is applied. A sensitivity analysis of the weights of the visibility and the straightening is carried out in order to assess different scenarios and to compare the existing TL path to the proposed ones. The method is applied to a case study where an old transmission line needs to be replaced by a new one and the local grid operator needs to identify feasible routes. A set of 30 routes is identified and most of them have a lower visibility that the existing path but, only some of them present a comparable complexity to be realized.

Generation and Validation of Spatial Distribution of Hourly Wind Speed Time-Series using Machine Learning

Wind resource assessment is a key aspect of wind farm planning since it allows to estimate the long term electricity production. Moreover, wind speed time-series at high resolution are helpful to estimate the temporal changes of the electricity generation and indispensable to design stand-alone systems, which are affected by the mismatch of supply and demand. In this work, we present a new generalized statistical methodology to generate the spatial distribution of wind speed time-series, using Switzerland as a case study. This research is based upon a machine learning model and demonstrates that statistical wind resource assessment can successfully be used for estimating wind speed time-series. In fact, this method is able to obtain reliable wind speed estimates and propagate all the sources of uncertainty (from the measurements to the mapping process) in an efficient way, i.e. minimizing computational time and load. This allows not only an accurate estimation, but the creation of precise confidence intervals to map the stochasticity of the wind resource for a particular site. The validation shows that machine learning can minimize the bias of the wind speed hourly estimates. Moreover, for each mapped location this method delivers not only the mean wind speed, but also its confidence interval, which are crucial data for planners.

Comparison of hourly and daily wind speed observations for the computation of Weibull parameters and power output

In this paper we compared Weibull distributions fitted using the hourly wind speed observations measured by weather stations of the UK Meteorological Office and daily averages measured by the National Oceanographic and Atmospheric Administration in the same locations. The measurements collected over 5 years (2009-2013) have been analysed. The shape and scale parameters and the mean wind speed have been compared. Finally, we considered a 3 MW wind turbine with 100m hub height to estimate and compare the long term electricity generation using the two datasets. The results show that, using daily averages, the mean wind speed values do not significantly vary. However, its standard deviation varies substantially, which is due to the smoothing of extreme wind measurements in the daily average dataset. When performing the long term estimate of the full-load hours and the electricity generation, the results show that error margins are low when using daily averages.

Validation of CM SAF SARAH solar radiation datasets for Switzerland

The CMSAF SARAH dataset is currently the only publicly solar radiation dataset with a spatial resolution of 0.05 degree and an hourly temporal resolution covering Switzerland over a time period of more than 30 years. The dataset has a higher bias in mountainous regions. Large parts of Switzerland, especially in the Jura and Alps, are mountainous. We compared the daily global and diffuse irradiance with measured irradiance data of ground stations at different elevations. The mean absolute bias increases with higher elevation. We investigated the correlation of the elevation of buildings with the error of the ground stations. Stations below 1000 meter elevation have an average mean daily bias of -0.18 ± 8.54 W/m2 representing 87% of Swiss building footprints.