Multiyear microgrid data from a research building in Tsukuba, Japan

Microgrids comprising renewable energy technologies are often modelled and optimised from a theoretical point of view. Verification of theoretical systems with data of actually implemented systems in the field rarely occurs in an open manner, especially on the intermediate scale of research buildings. To enable modelling of the actual microgrid performance of a research environment, we present a multiyear dataset of a microgrid with solar arrays and a battery. The main energy datasets comprise data per second supplemented by hourly solar irradiation data. These may be combined with data concerning the hourly electricity prices from the main grid and the low-electricity-price periods of national holidays. The level of detail of the data per second in combination with the hourly data in these datasets allows for a comparison to the efficiency and weather-parameter correlation of other renewable energy technologies, as well as forecasting future energy generation and consumption.

Microgrids comprising renewable energy technologies are often modelled and optimised from a theoretical point of view. Verification of theoretical systems with data of actually implemented systems in the field rarely occurs in an open manner, especially on the intermediate scale of research buildings. To enable modelling of the actual microgrid performance of a research environment, we present a multiyear dataset of a microgrid with solar arrays and a battery. The main energy datasets comprise data per second supplemented by hourly solar irradiation data. These may be combined with data concerning the hourly electricity prices from the main grid and the low-electricity-price periods of national holidays. The level of detail of the data per second in combination with the hourly data in these datasets allows for a comparison to the efficiency and weather-parameter correlation of other renewable energy technologies, as well as forecasting future energy generation and consumption.

Background & Summary
The current datasets consist of data related to a microgrid equipped with renewable energy systems and are gathered by an energy management system (EMS

Continually improve energy management
Although EMS can be applied to systems without microgrids, and neither EMS nor microgrids necessarily include renewable sources of energy, since 2000, there has been an evolving trend to manage energy generation and distribution by means of smart grids 1 . These have the advantages of improved energy reliability and security, shifted peak loads, increased efficiency, and reduced greenhouse gas emissions. In a similar trend, a more sustainable and integrated outlook on energy regulation has been taken by the Japan Smart Community Alliance 2 , who states that the main goals of a smart community are: These goals are much in line with the goals of the concerned microgrid, albeit that the current scale is on a building rather than community level.
Although there has been an increase in the number of studies regarding microgrids and smart grids applying renewable energy systems 3 , these often involve theoretical studies about optimisation 4-7 , whereas few studies perform field verification to test the actual system efficiency 8 , let alone on the building scale 9 . The building considered in the current study is a combined office and laboratory research building of five stories, which has been open since 2012. Actual building energy consumption datasets have recently been published for households 10 and office buildings 11 that are not equipped with or connected to renewable energy systems. The dataset provided through this study contains the data per second of the building energy consumption, battery charging/ discharging, and solar photovoltaic (PV) power generation; hourly data for solar irradiation; and related electricity prices per kWh. This allows for the examination of both the actual energy use of a research building with renewable energy systems, of which there are few studies published to date 9 , as well as the analysis of the efficiency of the PV and the battery.
The datasets may further be used as a basis for operational and planning studies of similar sites or for a comparison to freely available local weather data, and the values of each of the parameters may be forecast into the future. Individual array details are supplied to allow for a detailed efficiency analysis of different solar panel types. There are potential applications of these datasets in the fields of EMS design, energy-saving and demand forecasting, energy storage, computational optimization techniques and machine learning, economics and regulatory frameworks, and undoubtedly others.

Situational details
The individual datasets were collected from the NanoGREEN/WPI-MANA research building of the National Institute for Materials Science (NIMS). This is the first public building in Japan that has been equipped with a microgrid system consisting of four arrays of solar panels, a lead-acid battery, and an emergency backup generator (International Center for Materials Nanoarchitectonics (WPI-MANA), National Institute for Materials Science (NIMS), http://www.nims.go.jp/mana/about/building.html, 2018). In addition to this, a device is installed to measure solar irradiance. The purpose of the battery is three-fold: 1) to regulate frequency fluctuations within the microgrid, 2) to serve as backup energy supply during emergencies, and 3) to enable a peak cut of the energy demand and thereby reduce costs. The microgrid is incorporated within the Building Energy Management System (BEMS), which also records and regulates functions such as air conditioning, experimental gases, cooling and heating water, waste water, and electricity to 15 other research buildings on the same site. The microgrid was installed

Microgrid components
The main components of interest in the microgrid to this study are the four arrays of solar panels, a leadacid battery, and a pyranometer (see Fig. 1). There is also a backup power generator, which can be initiated during emergency power failures, although this has not occurred during the period of data recording. The details of the individual arrays can be found in Table 1. The azimuth of the building to which the solar arrays are mounted is 190°SW according to the building design drawing.
The lead-acid battery has a rated capacity of 326 kWh and a control maximum of 90 kW. The regular operation lies between 30 and 95% of the battery state of charge (SOC). The SOC is calibrated typically once a week, at night. The calibration is based on the voltage, which increases during charging and vice versa. When the voltage reaches the value of the specification during charging, the SOC is reset to 100%. Although the degradation of the battery may result in a different state at the same voltage, the same voltage value is used for calibration. As mentioned above, the battery has three functions: power quality assurance, backup power storage, and critical peak cut. The second function limits the minimum battery level of 30%; i.e., 30% of the capacity is saved as emergency backup. The last function is active only during the summer months (July-September), as peak demand usually occurs during those months. The functions of power quality assurance and emergency backup are active year-round and are of a higher priority than the function of critical peak cut to save both power and costs. Therefore, the operation of the battery is designed to have neither a large cost advantage nor burden for the consumer. Theoretically, the battery could serve a certain role in cost reduction due to its current operation of charging at night, when the electricity price is lower, and discharging during the peak hours of 13:00-16:00, when the price is higher. In general, the economic value of a constant emergency backup power supply is difficult to determine. The value of the lost load depends on many factors 12 , and in the setting of a research building, it is highly dependent on the type of experiments being run, the number of samples, and the treatment of the samples at the time of a blackout. Thus, calculating the value of the lost load would require additional information, which is beyond the scope of the current dataset publication.
The device to measure solar irradiance (pyranometer) is model LP PYRA 03. As indicated by the manufacturer 13 , "the measured irradiance (Global Irradiance) is the result of the sum of direct solar irradiance and of diffuse irradiance. LP PYRA 03 is a Second Class pyranometer". We believe the values recorded by the pyranometer may be less accurate than optimally possible for two reasons.
The first reason is related to its placement. The guidelines specify that the pyranometer be placed in a clear horizontal plane, without being hindered by any buildings, construction, trees, or obstructions. If this proves impossible, the pyranometer should be placed "where obstructions in the path of the sun from sunrise to sunset do not exceed 5 degrees of elevation", as "the presence of obstructions on the horizon line affects significantly the measurement of direct irradiance". This includes obstructions that can reflect light or produce shadows on the pyranometer, such as solar panels. We have found that the pyranometer has been placed directly at the bottom of Array 1, which consists of solar panels angled at 10°, and is therefore obstructed from the north side (see Fig. 1). The second reason lies in the lack of maintenance, as the guidelines recommend both cleaning the outer glass dome and annually calibrating the pyranometer to ensure its accuracy, yet neither of these has taken place over the last six years since its installation.

Data extraction
Microgrid data per second. Data extraction was performed on 26 and 27 April 2018. The data per second can be extracted through an extraction program compatible with systems running on Windows 7 and a Japanese OS. This program can extract a maximum of 16 parameters at a time. The data files created in this manner were CSV files, to be opened in Excel (Office 2010 edition, with Visual Basic 6.0 installed). Due to the limitations of Excel, the maximum number of rows a single file can contain is 1,048,576, meaning that the maximum number of days per file is limited to 12 (60 s × 60 min × 24 h × 12 d = 1,036,800 rows). We have extracted all data into 119 separate files for the period from 1 January 2015 to 24 April 2018.
Hourly solar irradiation data. Hourly solar irradiation data are collected automatically and transferred and stored into the BEMS. These are stored in daily files at 00:01 the next day. The hourly data is stored for up to one year and one month on a local server. Two datasets were made available through Seiko Solutions, Inc., who designed and manage the microgrid. Data extraction from the BEMS was performed on 26 and 27 April 2018, which led to a dataset that runs from 1 April 2017 to 24 April 2018. This was done by connecting a computer to the BEMS and transferring the available files (.txt format) through a LAN cable. A second dataset was made available through Seiko Solutions, Inc., and ran from 1 November 2015 to 16 March 2017. When combining these two available datasets, this has unfortunately left a data gap of two weeks from 17 March 2017 to 31 March 2017 in the hourly dataset. The parameter of interest in this data was solar irradiation.
Hourly electricity prices. The current prices are available on the Tokyo Electric Power Company (TEPCO) Holdings website 14 . An overview of past prices was obtained through the Facility Management Office of NIMS and added to the dataset (Data Citation 1 and Data Citation 2). Note that additional fees pertaining to the entire site and therefore other buildings are not included, such as fuel regulatory costs (adjustment charges) and basic connection and spare line fees.
National and additional holidays. Known national holidays were taken from the cabinet office of Japan (Table 2) 15 . In case a holiday falls on a Sunday, the next weekday becomes a holiday. Note that Mountain Day, 11 August, was first observed in 2016, and the Emperor's Birthday will change due to the scheduled abdication of the present Emperor in April 2019.

Code availability
Extraction of the data per second requires a computer with a Japanese OS using Windows 7 and requires Visual Basic 6.0, Excel 2007/2010, and the ability to connect to the BEMS through a LAN cable. The each consist of 18 of the larger NUOD062-067 panels and 12 of the smaller NUOD058-061 panels. b Arrays 2 and 3 are placed on opposite sides of a triangular shaped roof, angled towards the north and south (see Fig. 1). included code files (csv_sec) need to be installed in the folder with path "C:\user" (alternatively, the configuration settings can be adapted). Running the program "createSecCsv.exe" (type: application) allows the user to specify the time period from which to extract data. Upon pressing the button "create csv file" the data is output to an Excel sheet. Thirteen parameters from the 122 total available parameters can be selected for extraction by changing the last listed values in the file "CreateSecScv.ini" (type: configuration settings) named "device number". To enable understanding of the various parts of the code, written in Japanese, translations are given in parentheses in a separate file. Translations of the different device numbers, from which a selection of parameters can be made, are included in the dataset.

Data Records
The datasets in this study are available through the repositories of (Data Citation 1 and Data Citation 2). An overview of the datasets is given in Table 3. The microgrid data is available in both raw and cleaned formats for data per second. In addition, the data per second has been summarised into hourly data for easy comparison. To enable understanding of the different recorded parameters, their machine codes, and Japanese labels, Table 4 provides their translation/description. Two further datasets cover the electricity prices in Japanese Yen/kWh (Table 5), and a list of the holidays that influence the price in effect (Table 6). A final dataset consists of the code used to extract data per second from the microgrid to a stand-alone computer.

Technical Validation
The data per second is available for a total of 1,210 days; the hourly solar irradiation data is available for two periods of 502 and 389 days each (891 days combined). We have examined the data to identify and   remove erroneous data, which is inappropriate for use, resulting in the cleaned data sets. Note that the dates and times themselves have not been removed. The most voluminous data are the raw data per second, which combined have data on 12 parameters for 104,544,000 s, leading to 1,254,528,000 values in total (excluding headers).

Usage Notes
It may be necessary to convert the encoding of the CSV files to UTF-8-BOM to view Japanese characters in case the operating system does not recognize these correctly. We recommend opening the files with freely downloadable software such as Notepad++ Version 7.5.6 16 , converting the encoding (Encoding > Convert to UTF-8-BOM), and saving the file before opening in software such as Excel.
To calculate the economic efficiency of the individual solar arrays or the battery, we have included a dataset concerning the electricity price per kWh, which varies. For each hour, it is important to determine which of the four variable prices are in effect (Fig. 2). The feed-in tariff (FIT) surcharge is a required additional charge, which is always in effect during the recorded period. In addition, a solar surcharge under the renewable portfolio standards, which was in effect before the start of FIT in July 2012, was in effect until September 2014. Note that NIMS was exempted from the surcharge under renewable portfolio standards. The factor overruling all other remaining conditions is the occurrence of a Sunday or national holiday, during which the night time price is in effect. In addition, the same price applies to: 2 January, 3 January, 30 April, 1 May, 2 May, 30 December, and 31 December 16 . Next, it is important to determine the season. If it is summer, from 1 July to 30 September, two daytime prices are in effect. Between 8:00 and 13:00 and between 16:00 and 22:00, the summer daytime price is in effect. Between 13:00 and 16:00, the summer peak time price is in effect. When it is not summer, from 1 October to 30 June, the (regular) daytime price is in effect between 8:00 and 22:00. The night time price applies to all other times.
Finally, because the electricity prices change yearly as the contract is updated, it is important to determine the year. Figure 3 shows the past trend of prices. The dataset on electricity prices contains information on the entire period of operation since its initiation in 2013, to enable potential estimations of the hypothetical past and future trends.
For a comparison to local weather data, users are recommended to download the freely available hourly data measured at the Aerological Observatory in Tsukuba, situated approximately 1,500 meters away from the NanoGREEN/WPI-MANA building, from the Japan Meteorological Agency. Note that the website has a Japanese interface (Past weather data, Download (Japanese), Japan Meteorological Agency, http://www.data.jma.go.jp/gmd/risk/obsdl/index.php, 2018).