Articles

Interviews

Releases

New Products

Reviews

Editorial

Events

Sponsors

Site Search

Newsletters

Archives

Past Issues

Home

Editors

eDucation

Links

Software

Video Energy Twin at Haystack Connect virtual conference (Thu, May 6,12:25pm-1:40 pm EDT

Announcement of the joint Webinar - https://www.linkedin.com/feed/update/urn:li:activity:6800873785730981888/

Introduction

There has been a significant development in machine learning (ML) and integrating it into numerous fields of expertise in the past few years. As Sundar Pichai, the CEO of Google, said: "Machine learning is the future." Nowadays, ML also serves as a buzzword promising extraordinary results while increasing systems efficiency.

Since experts from the HVAC/Energy domain usually do not use Python, R, or Matlab most of the community does not use ML in everyday work. However, it does not mean that they cannot take advantage of ML and use its potential, bringing their work to a whole new level.

Meet Energy Twin (ET), state-of-art ML algorithms accessible to the SkySpark platform.

What is the Energy Twin?

ET is a machine learning SkySpark extension for energy consumption analysis, designed to efficiently monitor multiple buildings using artificial intelligence to identify problems and reveal the potential for future energy consumption savings and optimization. ET aims at hourly, or even 15 minutes, data modeling. This approach provides more valuable insight than aggregated daily, weekly or monthly data models.

ET was designed for energy optimization experts who usually do not have deep knowledge of artificial intelligence. The main objective behind ET is to combine state-of-the-art ML energy modeling techniques with a wide range of built-in SkySpark functions (Sparks, KPI, tariff engine, reporting, etc.). Thanks to smooth SkySpark integration, ET can be easily used without any coding or studying the mathematical background of energy modeling techniques.

Use Cases

ET is a versatile tool that can be used for various purposes, as described in this section.

Monitoring-based Commissioning (MBCx)

With monitoring-based commissioning, ET is used to detect significant deviations from normal behavior among a building portfolio and maximize the expert's time efficiency while minimizing the human-error effect.

Based on data, a model is identified using ET.

Fig1 - Example of ET model prediction (blue line) and measurement (red line) comparison. Prediction is influenced by the time of the week (see lower prediction during the weekend) and by weather (higher predicted values during high-temperature periods). 

Regardless of the building portfolio size, ET allows experts to supervise buildings more efficiently and identify anomalies or problematic buildings using Sparks and KPIs. The resulting ET analysis serves as an initiative for local investigations, which subsequently pinpoint the reason for abnormal energy consumption leading to solving the issue and preventing future problems. As a result, daily energy consumption can be lowered, ensuing financial savings. There are various reasons behind detected anomalies: local heaters always on, a non-stop operation of air doors due to manual regime override or a current meter with an incorrect multiplication factor. Some of them could remain undetected until a routine check-up is needed, resulting in unnecessary building operation costs.

Fig2 - The figure shows anomaly detection using the +20 kW threshold. There is a binary indication of timestamps in the bottom chart when the measurement is higher than the predicted value + 20 kW. The gray area in the top chart represents the estimate of the amount of avoidable energy.

M&V projects

SkySpark itself provides various features that are useful for M&V. Details and more information can be found on the SkyFoundry website.

On top of the standard SkySpark functionalities, ET adds ready-to-use modeling capabilities in adherence to the International Performance Measurement and Verification Protocol – IPMVP, ASHRAE Guideline 14 Measurement of Energy, Demand and Water Savings, and other M&V guidelines. Furthermore, ET evaluates the key statistical metrics that are defined by the guidelines. As a result, one can readily see if the identified model can be used as an M&V baseline model.

Identifying energy savings potential

Using ET analysis, we can identify the black sheep, a building with the most significant energy savings potential, in the building portfolio. Based on predictions and resulting KPIs, one can quantify potential savings. For a more detailed example, see ET Case study.

Peak demand prediction

Since ET can provide a 15-minutes prediction, it can be used in peak demand reduction scenarios. For example, using SkySpark’s built-in weather prediction, a model predicts energy consumption for the next days with 15-minutes sampling. This information can be used in a particular demand reduction (peak shaving) application.

Fig2 - The blue arrows show peak demand that was avoided thanks to the early warning based on the ET model. Local technicians were able to take preventive actions. Low priority equipment consumption was lower during the mid-day when full cooling power was needed.

ET is also compatible with the SkySpark tariff engine. This allows modeling future energy costs based on detailed pricing schemes and ET prediction.

Other possible uses of ET

ET was primarily designed for electrical energy consumption prediction; however, it can also provide a reliable prediction for other variables: water consumption or heat or gas (except for the heating systems with slow dynamics). Surprisingly, ET was also successfully used for modeling domestic hot water tank temperatures on a university campus.

ET background

The foundations of ET consist of a mathematical model with a similar structure as the TOWT model described in the paper: "Methods for Analyzing Electric Load Shape and its Variability" (Lawrence Berkeley National Laboratory Report LBNL-3713E, May 2010). A similar model is used in popular energy modeling packages used in M&V projects like RMV2.0, NMECR, or OpenEEMeter. More details about the model can be found in a separate document.

Developers can make use of an open API and use model prediction in custom SkySpark applications. ET makes all calculations On the Edge – only the standard SkySpark libraries are used. In other words, no cloud nor external libraries are needed. Additionally, there is no data sharing while using ET; your data are not stored anywhere or accessible for anyone else.

Conclusion

The possible practical benefits of applying artificial intelligence and machine learning-based applications in HVAC system operation are tremendous. Especially with large building portfolios with various types of facilities. Machine learning is an advanced technology, but one does not need a degree in Mathematics or Computer Science to use it. With the proper tools, such as the ET extension for SkySpark, the ML integration into everyday work is conveniently accessible. Some of the advantages that ET brings are:

·        maintaining high-standard services regardless of portfolio size,

·        maximizing efficiency while minimizing human-error effect,

·        focusing on the most severe anomalies by avoidable energy estimation

Using ML for analyzing a building portfolio makes it possible to maximize the experts' efficiency and take care of the mundane part of the job. Thanks to ET, the potential for energy savings is exploited, resulting in successful energy consumption savings and building operation costs.

Although ET does not fit all the needs, it is possible to make custom ET enhancements such as custom views or Sparks based on ET mode. There is also a possibility of creating a customized machine learning application in SkySpark. Each ET license includes 2 hours' worth of client support – to help with installation, configuration, modeling, or interpretation of the results.

For more information see the ET landing page https://et.mervis.info/  or sign-up for the joint Skyfoundry and Energy Twin webinar (June 3, 2021 at 11:30AM ET, email at info@skyfoundry.com for an invitation).

Authors bio

Jan Široký, PhD

As part of his doctoral studies, he focused on active fault detection and control with application to HVAC. During his doctoral studies, he completed an internship at ETH Zurich in the Automatic Control Laboratory. He has been working in the Energocentrum Plus company since 2003. He is the leader of a research team focusing on machine learning and automated analysis of HVAC/energy data. He is also leader of the Energy Twin team.

Jana Zázvorková

Jana Zázvorková is the Communications Manager of Energy Twin Team and currently finishing her Master's degree in Medical Electronics and Bioinformatics at the Czech Technical University in Prague.