Within the project ADA-EE a recommendation and decision support system for improving the energy efficiency of buildings was specified, implemented as a proof of concept solution and validated with a real demonstration building. The system specification of the advanced data analytics framework for energy efficiency includes use case description and requirements on advanced data analytics based on state of the art analysis and technology survey in the areas of CAFM, BEMS, data analytics, and decision support systems. Use cases are identified based on stakeholder and market needs, state of the art, and literature research. System requirements are identified and described which facilitates the defined use cases. Both functional and non-functional requirements were described and used for the implementation within the project.
The solution offers an automatic evaluation of monitoring data, prediction of future energy needs and prescribing measures to reduce energy usage. By continuously collecting real-time monitoring data, the algorithms can automatically improve prediction accuracy and prescribe better decision options. Methods such as data mining of monitoring data, forecasting and simulations are used to create decision recommendations for optimization of building energy efficiency. The system uses the up-to-date information collected from the property and data stored in the history of the property, together with weather data, as well as semantic data about the building (such as location, function, equipment type) which are analyzed to detect outliers or identify patterns and trends (descriptive analytics). Patterns are verified, correlations are detected and forecasts for the future are created by extrapolating the patterns (predictive analytics). The results of the previous steps are used to prescribe the optimal actions for improvement (prescriptive analytics). For this, first candidate actions are generated. They are then translated into a set of simulation input parameters. The parameters are used as an input into a baseline simulation, which includes automatic parametrization of the simulation model. The set of simulation results are the predicted potential consequences of the actions. Based on the results, a comparison is performed to recommend the actions with highest value (based on assessment criteria). The suggestions are formalized so that they can be used to generate automatic configuration files. For manual maintenance measures that should be done by maintenance personnel, manuals and human-readable instructions are automatically generated.
The system design enables flexible and modularized development. The specification enables an effective deployment of the system. Existing components that can be included are identified. Functional requirements such as expected inputs and outputs, data needs, etc., as well as the non-functional requirements were also defined. The structure of the system and connectivity between modules defines how data are acquired, stored and accessed as well as how communication between different modules takes place. Input as well as output interfaces are specified to decouple the abstract algorithms from their environment. The data analytics framework comprises descriptive, predictive and prescriptive algorithms.
Based on the requirements and the system specification, the interfaces to relevant external data sources were defined. These are an OPC UA and SQL-based monitoring data interface, a Web service-based weather data interface, and the semi-automatic semantic data interface. Subsequently, the infrastructure to store the data within the system was developed. This includes an OWL ontology for semantic information about the building and its automation systems, and an SQL database for monitoring and weather data. Monitoring data from the building automation system requires some pre-processing before it can be used for data analytics. Therefore, methods to remove data gaps, misalignments, and out of range values were implemented.
The descriptive data analytics module was developed to detect faults in the building automation system and to identify optimization potential via performance evaluation. Concrete use cases were defined for both tasks in order to support the specification of workflows and the implementation. Fault detection and performance evaluation rely on the same algorithms. An XML-based input format was specified for the description of how a specific fault or optimization potential can be detected. Together with monitoring data and semantic information from the internal databases, this information is used as input for the descriptive data analytics algorithms. An important goal for the implementation was that fault and optimization potential descriptions are not necessarily building-specific. They can be defined in a way such that they can also be used for other similar buildings with small or no adaptations. The results of the algorithms are then written to the semantic database.
The results of the performance evaluation were used for the generation of candidate actions. A candidate action is the adaptation of a single value within the configuration of a building automation system. For each identified optimization potential, all possibly suitable measures to use the potential, and therefore to increase the efficiency of the building, are fetched from the semantic database. Finally, all actions are combined to sets of candidate actions. Furthermore, weather parameters that can be used for data analysis and simulation were defined. Also, the forecasting models for sun radiation and cloud cover were improved.
For the mapping of candidate actions which consists of translation of actions into simulation parameters as well as running a baseline simulation using a geometry model and an HVAC model, existing tools such as SketchUp and TRNSYS were integrated using scripts. An energy forecasting mechanism was designed based on the baseline simulation and an integrated weather forecasting model and monitoring data. The workflow was integrated using a combination of an MQTT broker and an FTP server. The output files of the energy forecasting are adapted to the assessment methodology. A method for automatic improvement of the building models was proposed and implemented using GenOpt.
Existing prescriptive analytics algorithms including their requirements and challenges were evaluated in order to better implement the recommendation assessment process. The aim of the assessment process is selection of the potentially best candidate action set to improve energy efficiency and decrease occupants' dissatisfaction. This includes definition of key performance indicators (KPIs) as assessment criteria. The current implementation includes total primary energy, thermal comfort and air quality as criteria. For further processing, the selected candidate action set is formalized as a recommendation. Results of the assessment process can deliver useful feedback for the generation of candidate actions in the data preparation and data description module. Two concepts for such a feedback loop are considered, short-term reduction and long-term optimization of the candidate action generation. Different weather model resolution and accuracy is analysed using a cost/benefit analysis.
Formalized selected recommendations form the basis for the generation of configuration and decision support reports. During the definition of the formalization process, it was already taken into account that suggested actions must be interpreted to set new values in the BAS and for that a common configuration model was specified. The model is represented as an OWL graph which is contained in the ontology that is used in all phases of the developed system. It contains the properties respectively set points which must be changed, their location in the building, new values, effects on the building, as well as additional information like date and time of the detection of optimization potential.
Multiple state-of-the-art building automation technologies like KNX, BACnet, and LON were investigated in order to define a mapping between the common model and these popular standards. Particular attention was paid to configuration and commissioning procedures. It turned out that the specification of a fixed and exhaustive mapping is not feasible, because of several reasons which are associated with the way of how device configuration parameters are standardized respectively not standardized for these technologies. Hence, it was decided to use a state-of-the-art integration technology, namely OPC UA, to enable the automatic adoption of BAS configurations. An XSD schema was specified for the definition of mappings between control services in the ontology and OPC UA node ids which address the corresponding parameters and set points in the BAS.
An OPC UA client was developed based on the mentioned conclusions. It uses the common configuration model and the defined mapping as input and writes new values to an OPC UA server which is responsible for the propagation of the values into the actual automation system. For purposes of testing and development, OPC UA servers were implemented and populated with data variables. Such servers are also available for most of the modern BASs.
Besides the automatic transfer of configuration changes into the BAS, human-readable reports are generated for the facility management. Two kinds of reports are defined and designed. Fault reports inform facility managers about problems which were detected in the building during the fault detection process. Optimization reports, on the other hand, describe recognized optimization potential, configuration changes to use these potentials, and expected effects on comfort respectively energy consumption in the building. The generation of these reports relies on XML related technologies like DOM, XSLT, and XSL-FO. The outputs of the process are PDF files which can be distributed to the facility management.
A handbook was written describing the deployment process and usage of the advanced data analytics framework including results and lessons learned from the evaluation. As part of this handbook, first, the preparation for deployment is described including data collection, modelling preparation, and assessment and decision support analysis. Then, the deployment process is explained: This includes setup of internal databases and external interfaces, setup of data analysis and candidate action generation, setup of the simulation environment, configuration of the assessment criteria, setup of the generation of human readable reports as well as automated BAS configuration updates (if required) and configuration of a scheduler for continuous operation of the framework. Finally, the results of the evaluation of the data analytics framework as well as lessons learned are presented.
The market potential of advanced data analytics for energy efficiency is analyzed based on a rough estimation of the additional investment and the potential effects of the improvements using a cost/benefit analysis. The deployment itself requires additional investment costs (e.g. for additional equipment or modeling of the building and its systems) and only seen through the savings the return of investment period based on the assumptions is somewhere between 5 and 10 years. However, the indirect effect of the occupants' comfort and company's image improvement, which are difficult to measure, is another major benefit and should be a driving factor for deployment.