In-Memory Technology for Smart Wind Farm Control

Viktor Dmitriyev, Jorge Marx Gómez and Manuel Osmers

With the growth of the economy, population of countries and newly appeared habits of humankind (e.g. mobile devices) energy consumption is continuously increasing. During the last two centuries people all over the word developed very robust techniques for energy production. However, in order to maintain newly appeared tools, the new methods and processes are required. This article demonstrates the ability of the new database technology such as in-memory computing [1] to be able to work with huge amounts of data generated by wind turbines in order to reduce maintenance costs through more detailed predictions mechanisms.

Developed techniques for energy production are quite mature but their major drawback is that such techniques are not sustainable enough. In most of the cases, techniques like burning coal or oil to produce energy do not only bring a significant damage to the environment but also becoming more and more expensive. Understanding this aspect gives a big push to the development of the renewable-energy sector. One of the most promising directions in the renewable-energy sector is wind energy. The total amount of wind energy turbines raised significantly during the last years. Therefor governments and companies investing more and more money in research, development and infrastructure. 

As always, new domains bring completely new challenges along with previously existing. One of the classical challenges that has always existed and will exist due to the universe’s entropy is maintenance of existing infrastructure. The major aim of the Smart Wind Farm Control project is to solve issues of increased costs and efforts needed for maintenance wind turbines located in offshore farms and clearly demonstrate applicability and feasibility of the in-memory solution. The objective is to capture the whole data traffic coming from wind turbines, detect error chains by using data mining methods and use gathered knowledge for proactive maintenance. On one side, the development of the Smart Wind Farm Control project is inspired by the companies dealing with offshore wind farm parks on daily basis. On the other side, there is a great possibility to adopt ideas and experience for onshore wind farms.

Point of View

Wind farm can be defined as a set of wind turbines combined by the location principle in one single energy grid. One single wind farm consists out of multiple wind turbines. The total number of turbines used in one single wind farm heavily depends on the architecture and the applicability purpose (e.g. factory, civil sector, amount of investment, wind speed, etc.). The reason to combine wind turbines together in one wind farm is to increase the overall energy production rate and decrease infrastructure costs overheads. 

Mainly, wind farms can be classified into two groups: (a) onshore (located on the land) and (b) offshore (located on the sea). There is a long list with differences between onshore and offshore wind farms. While onshore wind farms are relatively easy to maintain due to the easier logistic, offshore wind farms cause high maintenance costs. There is a variety of reasons for this: (a) restricted means of transportation; (b) dependency on meteorological conditions; (c) more complex supply chain, due to the lower amount of offshore wind farms installation. The main ambition of the project members is to have such technical solutions, that it should not only offer an ability to collect and store data but it also should offer complex analysis like building predictions and real-time monitoring/reporting which are beyond of the scope of general purpose systems. Besides generating energy, each single wind turbine within the wind farm generates telemetry data illustrating process of energy production. Latest modifications of wind turbines contain 400 and even more sensors, which are able to emit measures on second basis.

In the era when most of the decisions on all levels (strategic, tactical and operational) are heavily influenced by the knowledge obtained from careful analysing data, when data-centric models are bringing new inspiration to a business model by changing them, when business started to treat data as a "new oil" it’s quite straight forward to use data that is generated by wind turbines to enhance usage experience, save expenses and reduce maintenance costs. However, abnormal amounts of data generated on second basis by a single wind turbine need a specific technical solution. So in order to meet the ambition of the project, SAP HANA [2] was selected as most appropriate technical solution with in-memory data storage and out-of-the box analytical capabilities.


The Smart Wind Farm Control project is primary focusing on maintaining offshore wind farms. The objective is to capture the whole data traffic coming from wind turbines and to detect error chains by using data mining methods and use gathered knowledge for proactive maintenance. Most of the current analyses are performed on pre-aggregated data. However one of the main hypotheses of the project is that looking at data on more granular level can lead to more precise predictions and more accurate proactive maintenance. An advantage is that the 400 sensors of the wind turbine are already delivering data on second basis. Assuming that each single sensor is able to produce 10 bytes of information per second the total amount of data in one year per one single wind turbine results into approximately 117 GB (10 Bytes x 400 Sensors x 60 Seconds x 60 Minutes x 24 Hours x 365 Days ~ 117 GB). However by utilisation of in-memory database such as SAP HANA the creation of pre-aggregates and building pre-defined OLAP cubes and can be omitted instead. Then aggregations can be done “on the fly” due to the in-memory nature of the main data storage. Hence, analysis procedure can be more flexible and can be performed directly on the data taking into account any level of granularity. Besides working with on-demand aggregations, it is a great possibility while working with SAP HANA to use its native R-integration capabilities. Such integration helps to run much more comprehensive analysis on the data in order to get better results on predictions.

The final system solution should offer following capabilities: (a) Proactive Management System; (b) Precise Forecasts for Maintenance Periods; (c) On-Demand Statistic for Physical Investigations. The Proactive Management System capability is needed to evaluate all relevant physical values, which are provided by the offshore wind park. The main task of the proactive maintenance is to calculate the average remaining life expectancy. Furthermore, it should provide an automated error detection and error classification unit. Real-time monitoring and reporting should be based on a dataset containing 400 records per second per turbine. The Precise Forecasts for Maintenance Periods capability is focusing on the turbine maintenance, the main goal of which to forecast lifetime estimation and breakdowns. Where forecast reports and pre-alerting for all turbine components should be created automatically. It should be able to generate these reports and alerting using weather data, resource data, operational data and maintenance history data. The On-Demand Statistic for Physical Investigations capability is an ability to provide and execute more complex analyses on a larger dataset in the shorter period of time. Thus, faster responses create a possibility to improve the investigation operation needed for problem solving. 

The project was done in cooperation with multiple partners. The Hasso Plattner Institute (HPI Future SOC Lab) partner not only offered technical landscape facilities including SAP HANA but also provided the project with SAP HANA related expertise which was very helpful in achieving better results. Other industrial partner supported the project with real production data from offshore wind farms and also provided domain expertise.


The final system solution is divided into three layers connected to each other through special interfaces. As it is shown in Figure 1, the architecture is divided into three following layers: (a) Extract, Transform, Load (ETL); (b) Data Storage and Data Mining; (c) Reporting.

The Extract, Transform, Load (ETL) layer generally addresses data collection, data cleansing and transformation. Within the Smart Wind Farm Control project, the data is available in the form of historical and generated wind turbine data. In addition, continues data collection, such as weather conditions or maintenance data can be established. For a potential productive use a continuous data stream of various wind turbines may be another data source. On the early stage of the project the Pentaho Data Integration CE (Kettle) was used as a main ETL tool that allowed cleaning up of the historical and supplementary data and transforming them into the correct data model. An own home-solution SWF Toolbox was implemented afterwards in order to simulates real streaming nature of data generates by wind turbines.

Figure1: Architecture Overview.

The Data Storage and Data Mining layer consists out of two components: (a) SAP HANA and (b) R/Reserve. The role of the SAP HANA subcomponent of the architecture is to act as a single point of data storage. The database model of which includes eight tables with different numbers of attributes which are able to gather all kinds of incoming data, related to the maintenance of offshore wind turbines. The role of the R subcomponent is to perform data mining tasks. The data mining results are dynamically sent to an email address as an alert (proactive maintenance) and also store inside SAP HANA.

The Reporting layer is presenting data to the end users in a graphical way. Microsoft Excel is used for rapid and lightweight analysis and reporting in form of charts, tables and pivot tables. Furthermore a web application based on SAP UI5 framework for analysis and reporting of wind turbines was developed. The web application is designed to provide an overview of the major functional areas such as monitoring, log viewing, reporting and data mining. The reason to use both support with almost the same functionality in the web part of the reporting is to demonstrate various application server capabilities of SAP HANA and also offer additional functionality which is not available (or hardly achievable) in Microsoft Excel.

Despite being a proof of concept the Smart Wind Farm Control projects consist out of multiple noticeable features such as: (a) advanced analytics on demand which is based on SAP HANA in-memory database and it is analytical capability; (b) integration with well-know and widely acceptable front-end analytical tools such as Microsoft Excel; (c) ability to offer reporting and analysis results ubiquitously through modern web browsers, supported by SAP UI5.

Discussion and future directions

The proposed solution can be treated as a base platform for future projects. The project’s principal architecture design with SAP HANA as main data storage and processing unit allows enhancements that can be achieved by building more accurate predictive models. The accuracy of the future predictive models can be driven by using better feature selection techniques or by utilizing state-of-the-art and cutting-edge predictive algorithms offered by SAP HANA’s Predictive Analytics Library (PAL) [3]. Maintenance data has to be treated carefully due to free text fields to describe the maintenance allowing spelling errors such as mixed abbreviations, missing blank characters as well as use of upper and lower case. 

The Smart Wind Farm Control project got positive feedback from industrial partners. The companies are quite interested in further enhancements of the topic and we are working with them closely. 




[1] Plattner, H.; Zeier, A. (2012). In-Memory data management: Technology and applications. Springer Science & Business Media.
[2] Färber, F.; Cha, S. K.; Primsch, J.; Bornhövd, C.; Sigg, S.; Lehner, W.: SAP HANA database: data management for modern business applications. ACM Sigmod Record, 40(4), 45-51. 2012.
[3] SAP, AG SAP. SAP HANA Predictive Analysis Library (PAL) Reference, 2013.