Filling Gaps in Micro Grid by Method Based on Empirical Orthogonal Functions

Abstract

An analysis of Micro Grid system performance requires both meteorological and electrical data for the assessment period. However, actual in-field data acquisition is rarely 100%, often resulting in a significant amount of incomplete datasets for performance assessment. These gaps, if not taken into account, may add noticeable bias in yield assessment and thus estimations of the lacking data need to be made. Approaches of back-filling the required data is given and validated here. This paper presents a strategy to back-fill data with good accuracy for both short and long term periods, while taking into account weather as well as system performance variations. Cases of data loss are identified. The first case is that of missing meteorological datasets, while electrical readings are available. This case is met in most small systems, either domestic or commercial, where installers reduce the cost by omitting the meteorological sensors. The second case is that of the electrical monitoring system being interrupted. The third case is a failure of both monitoring sub-systems, which could be due to communication or hardware failures. The last two cases are often met in the majority of solar farms. The application of the Heisenberg uncertainty principle when operating a Micro Grid indicates the need to predict the amount of energy that can be obtained from the station at the next observation interval. For the implementation of predictive control, it is necessary to predict the change in the illumination of solar panels, provided the cloud passes over their plane. Method based on empirical orthogonal functions reconstructs missing data using empirical orthogonal functions, derived from the original data. While EOFs in a complete dataset would typically be calculated using singular value decomposition, the presence of missing data requires an iterative approach. The method allows for the estimation of missing values and full EOFs by first inserting mean values into the missing portions of the dataset and then calculating the EOFs. Because the resulting spatial EOFs and the time series of their magnitudes reconstruct the original data, a truncated version of the original dataset can be generated, using only as many EOFs as are deemed significant through validation. This provides an improved estimate of the missing information over simply inserting mean values, because the small-variance (i.e., noise) EOFs have been removed. The city of Bottrop, Germany, has been selected as the research object. In this area, most roofs have solar panels. The schematic representation of the part of the district with known and missing data, as well as the direction of the projection of the cloud, is given. The data matrix of the illumination is provided, provided that the sensors are installed in such a way that the data are obtained in each cell, the matrix with the data absent due to the absence of sensors and the matrix with the restored data. It is shown that if the coefficient of transparency of the atmosphere is changed according to the sinusoidal law and when the third degree polynomial is used to restore it, the accuracy is no more than 1%.

Ref. 10, fig. 6.