Multivariate Wind Speed Forecasting with LSTMs for Wind Farm Performance Estimation

Abstract

One of the most important renewable energy sources today is wind power generation. However, this energy source depends on the flow of air in the area where the wind generators are installed, and as with many other renewable resources, this is a natural resource whose magnitude is not directly controlled by man. The prediction of wind speed then becomes a key problem when we want to project the energy performance of a wind farm. The behaviour of the wind, as a climatic variable, can be estimated from various atmospheric parameters such as temperature, humidity and air pressure. In this paper we propose a multivariate wind speed estimation model from the history of these atmospheric parameters using a deep neural network. The performance of the model is then evaluated against the same historical data, which produces a fairly small error. The code is implemented in Keras deep learning library with TensorFlow backend. Keyword-Atmospheric parameters, Deep learning, Forecasting, Kinematics, Multivariate model, Wind speed I. INTRODUCTION Many developing countries such as Colombia must focus their resources correctly in order to support the development of their individuals [1]. Electricity is a fundamental element for the well-being of individuals and the industrial development of a country, therefore it is also widely used as an indicator of development. Colombia has many areas of the country without electricity service [2]. This is mainly due to the costs of bringing the traditional power system to remote areas with low population density. However, these areas are characterized by own resources that can be redirected to favor the population. There are areas of the country with a very good wind resource, which can be used to generate electricity as an energy solution in specific areas [3, 4, 5]. However, the implementation of this strategy entails many challenges. One of them is related to the projection of wind energy available in the areas. The planning, projection and operation of a wind power plant needs to quantify a priori the available wind resource [6]. This is also important in cases of interconnection with the traditional power grid. However, predicting wind speed is a fairly complex problem due to its fluctuating behaviour and the number of variables that affect it [7, 8]. Most of the strategies currently used to predict wind speed have faults, and even sometimes incur large errors. Among the conventional strategies used to estimate wind speed is the Numerical Weather Prediction (NWP) method [9]. This method consists in an interpolation of values from known data in nearby places. In order to improve the behaviour of these methods a correction strategy is usually used, one of the best results are recurrent neural networks given their natural ability to learn time series [10]. Another conventional method applied uses the Persistence model [11, 12]. In this case the method is able to provide very good short-term values due to its short-term prediction capability. However, when we want to make projections with a ...