Demand Flexibility

s  Demand Flexibility

Buildings configured to operate to respond to price signals or other mechanisms (eg wholesale demand response management) can provide demand flexibility in building-to-grid applications. Occupant comfort or production line reliability may be over-riding factors. Learning algorithms use a reward mechanism; the reward (the “definition of good”) is defined by the user story of the building owner and tenants.

post thumb


Demand Flexibility Workshop

In the Demand Flexibility Workshop we work together to discover what opportunities for demand flexibility exist; both for gas and electricity for your building, site, campus or portfolio.

Determine what is important for your operations; sacrificing occupant comfort is rarely a desired outcome. However, there are plenty of control optimisation strategies that can be explored that improve occupant comfort while reducing electrical demand and consumption.

Model the impact of electrification - layer existing gas usage on historical electricity consumption data.

NEM12 Electricity Metering Data

We prefer to use the AEMO/AER standard NEM12 metering data format as it supplies us with active and reactive power, from which we calculate maximum demand and power factor (using the same calculation methods as the retailers).

We can also accept CSV and other custom data formats - but usually data quality is not as good as NEM12, thereby can cause additional costs in terms of time to be accrued to effectively process the data.

Rapid Cost Assessment

Capability Overview

Find the optimum mix of energy generation, storage and demand using a high-level data-driven approach. Load the ground truth data from NEM12 or CSV, then ‘mutate’ or modify historical data to simulate what-if addition/subtraction of energy related systems including:

System Sizing

  • solar PV system size increase;
  • battery peak lopping & arbitrage; and
  • battery FCAS.

Control Strategies

  • HVAC load-shifting (pre-heat/pre-cool);
  • ventilation night-purge;
  • energy consumption reduction/increase; and
  • demand reduction/increase.
  • adding an embedded network;

Financial Impacts

  • load profiles (ground truth to what-if);
  • discount cash rate;
  • inflation;
  • cap-ex (incl equipment replacement costs);
  • op-ex;
  • tariff change (network/retail);
  • annual credits; and
  • power factor.

Meaningful Reports for Decision Makers

Compare what-if scenarios to the ground-truth using:

  • 15 & 50 year Benefit Cost Ratio (BCR);
  • 15 & 50 year Net Present Value (NPV); and
  • 50-year discount cash flow.

Find Out More About Deep Energy

Design, Documentation & Tendering

BEM System Analysis

For more complex requirements, a model-driven approach will assist validating assumptions used in the Rapid Cost Assessment process.

A Building System Analysis summary report, incorporating the above requirements. The report includes:

  • building name & location;
  • modelling software used;
  • occupancy type (apartment, aged care, over 50s, commercial, retail, supermarket, office et cetera);
  • gross floor area/net lettable area;
  • energy use summary report, broken down to different energy sources and energy uses, including generation and storage (thermal & chemical);
  • energy use intensity (EUI);
  • electrical maximum demand (and gas, if a component);
  • mechanical, electrical, building envelope and occupancy schedules;
  • window-to-wall ratio, fenestration-to-wall ratio (including doors);
  • air leakage rate;
  • lighting level intensity and use cases to calculate lighting load;
  • EEO section, as detailed below;
  • Rapid Cost Assessment, as detailed below; and
  • provide a summary and list of recommendations along with proof of data-driven decision making.

Focus on Energy Efficiency Opportunities (EEO) so that we:

  • identify and list a range of EEOs from the design development process;
  • nominate three EEOs to run as what-if scenarios in the RCA; and
  • EEOs that focus on operational modes and consequent reduction in equipment life-cycle replacement costs will be favoured for analysis.

Feed results back into Rapid Cost Assessment to:

  • quantify the financial benefit versus the reference design using NPV and BCR;
  • incorporate capex and opex costs, including equipment replacement based figures from BEM/BSA;
  • provide a full cash summary over 50 years;
  • GHG emissions and GHG intensity from grid (interval emissions intensity from scope 2 sources); and
  • provide comparative analysis of three EEOs against the baseline design incorporating capex, opex and life-cycle replacement costs.

Design & Documentation

Our team has extensive experience developing engineering briefs, reference designs, schematics and specifications for data-driven smart building projects. Our focus is to remove ambiguity and reduce risk of high variability in costs from tender responses.


We assist clients with preparing and evaluating tenders for building energy systems and services.

CSIRO Smart Building Data Clearing House (CSIRO DCH)

The CSIRO platform is pre-release and is made available to select R&D projects via Buildings Evolved.


  1. Single place for all your data: Collect all your building data in one place from disparate sources while retaining ownership irrespective of service providers.
  2. Get insights from your data: Discover new insights from your data by leveraging our smart building models.
  3. Apps to unlock your data: Deploy applications through DCH to supercharge your operations and analytics. App providers get access to building data without installing new hardware
  4. Unlock opportunity for advanced optimisation - The smart building data clearing house provides a unique opportunity for data providers and innovators to come together to improve the operation of the grid.

Where there are many vendors that have created vertically integrated platforms, CSIRO set out to create a horizontally integrated platform in recognition that stock markets operate on a similar precept. The data clearing house is intended to facilitate a marketplace of data and applications that provide value to these data. The owner of the data can share to vendors, consultants or other parties and control access revocation when respective works come to completion.

Optimisation Apps

Traditional approaches call for development of complex physics and building thermal models in order to optimise the building against. Open AI and reinforcement learning provides a new approach – data-driven rather than model-driven, supported by a simpler ontology.

Optimisation algorithms are typically executed in the CSIRO Data Clearing House on a schedule, using available sensor data and external datasets to determine the best operational strategy for the intervening period (user defined). Long term forecasts provide forward projections of optimum control. Unforeseen events (e.g. a generator tripping) are monitored to see if real-time actions are required, for example a price driven demand response mechanism.

A model-driven approach requires detailed building physics and simulated building controls to create a baseline – a theoretical level of performance that the building will be able to attain. LEED certifications in the United States expect a detailed model to underpin the design and delivery process. Real world building performance is compared to the simulation. The building, through upgrades and maintenance should maintain or better the performance indicated in the simulation. While this is a recommended approach with the increasing complexity of energy systems in the built environment, it is not always achievable, particularly on existing sites, or large existing portfolios of property.

A data-driven approach does not require detailed physics models in order to operate. Rather, a mechanism uses AI to “test” what happens with any given permutation or combination of variables within valid real-world physical constraints. A “reward” mechanism tells the algorithm if it did good or bad in with that particular combination of variables. It keeps trying until it finds the optimum mix of controls to achieve whatever the stated control strategy is designed to achieve. Data-driven approaches use neural networks to create an inference model. Learning algorithms use the inference model to determine what the next “guess” is. Rewards are issued when the system gets it right.

There are several levels of control possible:

  1. Rules-based control the set point and run on a schedule (smarter, but barely);
  2. ML based control of HVAC mode, fan speed as well as set-point, but still schedule on/off; or
  3. **Fu For example, if you have a sunny day, it is highly likely that electricity spot prices will go negative during the day. Then there is the question of how we induce demand artificially to take advantage of cheap prices (balancing the cost factors of the network tariff and maximum demand) – because, again, this is what the grid wants in that scenario.

External Data Sources

Data from the Bureau of Meteorology, the Australian Market Operator and CSIRO Energy are used to provide context for control optimisation forecasting. Aligning the forecasts of weather and electricity prices provides a powerful context from within to make decisions about control strategies. Price signals from actual wholesale spot price of electricity can be used to interrupt programmed control logic - price driven wholesale demand response, rather than using the formal mechanisms designed for very large consumers of electricity.

Price goes over $600/MWh? Time to do some load-shedding, perhaps. For example, if you have a sunny day, it is highly likely that electricity spot prices will go negative during the day. Then there is the question of how we induce demand artificially to take advantage of cheap prices (balancing the cost factors of the network tariff and maximum demand) – because, again, this is what the grid wants in that scenario.