Assessment of wind energy resources using artificial neural networks – case study at Łódź Hills

BULLETIN OF THE POLISH ACADEMY OF SCIENCES TECHNICAL SCIENCES, Vol. 67, No. 1, 2019 DOI: 10.24425/bpas.2019.127340 Assessment of wind energy resources using artificial neural networks – Case study at Łódź Hills R. KORUPCZYŃSKI 1* and J. TRAJER2 1 2 Electrotechnical Institute, Pożaryskiego St. 28, 04-703 Warsaw, Poland Warsaw University of Life Sciences – SGGW, Faculty of Production Engineering, Nowoursynowska St. 164, 02-787 Warsaw, Poland Abstract. The aim of this paper is to answer the question: Are the Łódź Hills useful for electrical energy production from wind energy or not? Due to access to short-term data related to wind measurements (the period of 2008 and 2009) from a local meteorological station, the measure – correlate – predict approach have been applied. Long-term (1979‒2016) reference data were obtained from ECWMF ERA-40 Reanalysis. Artificial neural networks were used to calculate predicted wind speed. The obtained average wind speed and wind power density was 4.21 ms –1 and 70 Wm –1, respectively, at 10 m above ground level (5.51 ms –1, 170 Wm –1 at 50 m). From the point of view of Polish wind conditions, Łódź Hills may be considered useful for wind power engineering. Key words: wind speed, artificial neural network, wind resource, measure-correlate-predict. 1. Introduction Renewable sources of energy have been increasingly used over the recent years. Fossil fuels are running out, and some CO2 emission limits are introduced, hence there is a strong need for looking for other energy sources worldwide, as well as in Poland [1]. One of free renewable resources of energy is wind. Wind energy might be converted by a wind turbine into electrical energy. The nominal power of a typical wind turbine is not very high, it is equal to about 2 to 3 MW [2]. Often, many wind turbines are concentrated in one place and form a wind farm. The total cost of a typical farm is quite high (i.e. 200 MW land-based, USA): 1690 kW of installed power [3]. However, there is no other option, since, according to the European Union Directive, renewable energy share in the total energy has to amount to 20% by 2020 [4]. Efficient use of wind energy requires choosing a site with the highest possible wind speeds, as temporary, electrical power of a turbine is proportional to the cube of wind speed [5]. But the wind speed is too variable parameter. It can be described mathematically by means of statistical methods [6]. The value of wind speed is determined at the given site by the method of real measurement. There is a potential risk of measurement campaign execution at a period with higher than usual wind speeds. Hence, the measurements data should be available in long-term period of time, i.e. 10 [7] or 20‒25 years [8]. This kind of data is usually held by national meteorological services. From the wind farm developer point of view, a waiting time of investment launching may not be too long. The measurements period should be as short as possible: the standard is one *e-mail: Manuscript submitted 2018-03-27, revised 2018-09-17, initially accepted for publication 2018-09-21, published in February 2019. Bull. Pol. Ac.: Tech. 67(1) 2019 year [9]. But this is contradictory to what was stated above that the period should be at least 10 years. A solution to the problem is the application of Measure – Correlate – Predict method (MCP) [10]. It requires two shortterm data sets: one from meteorological station at given place, and the other one, so called reference dataset. The sets (here for the period of 2 years) are the basis of artificial neural network model training. After the training, the model is used to predict wind speed time series at a given place with the third, long-term reference dataset. The artificial neural network model is better than a mathematical function, because it better describes the correlation between the datasets being analysed, including many influencing factors. After positive verification of the model, wind energy resources at the given place can be evaluated considering the following parameters: mean wind speed, wind speed distribution and wind power at a given height, usually equal to the proposed wind turbine height. The aim of this paper is to create the prediction model of wind resources at Łódź Hills. After the model application, wind energy resources will be determined, as well as usability of subject location for wind power engineering. 2. Location Łódź Hills (Heights) are the southern part of the Mazovian Lowland. In the physico-geographical regionalization of Poland this Mezoregion has the number of 318.82 [11]. The landscape is made up of rolling upland. A few cities are located in this region, such as: Łódź (the largest), Zgierz, Brzeziny, Stryków and Rzgów. In the northern part of the region, Łódź Hills Landscape Park is located. Apart from the area of Łódź, where industrialised landscape dominates, the rest of the region has agricultural character. Hence, this terrain may be convenient for wind energy developments. As stated in [12], Łódź Hills are located 115 R. Korupczyński and J. Trajer on the border of the beneficial and the highly beneficial wind zones, according to the classification published by the Institute of Meteorology and Water Management. In the whole Łódź Province, there are wind farms, i.e. at the Kamieńsk Mountain, in the vicinity of the cities of Głuchów and Słupia. (ECMWF). The time period of available data is quite long: 1979 to present (2016). In this paper, reanalysis plays a role of reference meteorological station for the MCP method. 3. Measurement station 5.1. Measure – Correlate – Predict Method. The Measure – Correlate – Predict method was first described by Derrick [18]. In this method, the required dataset consists of two parts: one-year measurements data from the given site and multi-year data from a local meteorological station. Beside the wind speed, the wind direction can be used [19]. The period of time, in which the measurements were made at the given site, has to be included in the long-term dataset from the reference meteorological station. Next, the correlation function between the data from the reference station and the site is calculated, based on the common period of time in both datasets. The last step involves using the correlation function for the calculation of long-term prediction at the given site [20]. Long-term data from the reference station are used as input data. The correlation can be determined using different methods, i.e.: linear regression [19‒21], variance ratio [19], artificial neural network [22, 23], support vector machines [24]. A comprehensive and detailed review of many MCP variants is included in [25]. The accuracy of the MCP method depends on various factors and changes at various locations. Mean wind speed prediction error equal 3‒10% have been reported [20, 26]. From the point of view of energy yield, error of approx. 4% has been mentioned in [27]. In this paper, the d (...truncated)