Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Data driven techniques to model thermal and electrical demand of energy hub
Optimization of a multi-energy system rely on future demand profiles for the energy and heating demand of the hubs. The goal of this project is to evaluate data driven approaches to model hub demand patterns as opposed to first principal models that can be difficult or costly to develop/maintain.
Keywords: Optimization, Large-scale systems, Energy hub, Data driven modelling, energy demand forecasting,
machine learning, model predictive control, multi-energy systems, distributed control, MPC, model predictive control,
Energy hubs are multi-carrier energy systems that incorporate conversion, storage and network technologies of different energy carriers such as electrical, thermal and/or chemical processes that follow individual demand and supply patterns under distinct time scales. Energy hubs facilitate easier integration of intermittent renewable energy sources, greater flexibility and higher efficiency of the overall system. Furthermore, the joint operational optimization of multiple hubs in the network that considers the energy trading between them further increases flexibility and RES utility, and reduces the overall operational cost of the system. The resulting system is also better equipped to meet challenges such as demand, generation and price uncertainty.
A plug and play framework that models a network of energy hubs has been designed along with a novel distributed control strategy for the multi-objective optimization of the network. The cooperative operation of the network and control strategies rely on future demand profiles for the energy and heating demand of the hubs based on the model of the hub. The ability to predict a hubs energy consumption, using a variety of performance metrics is crucial for future energy scenario planning. Data Driven modelling approaches are increasingly utilized since first principles models may be costly to develop, difficult to maintain and in some cases, first principle identification methods cannot be effectively employed. The increased quantity and quality of data collected facilitates the utilization of data-driven approaches, thereby realizing the potential for energy prediction as a complementary or alternative option to the more traditional physics based approaches and reducing the reliance on first principle models. These methods can also be tailored to factor in the seasonal and price dependency of a load. Data driven methods such as Gaussian process regression are also flexible due to its inherent ability to describe uncertainty in the prediction making it a suitable candidate for stochastic optimization and provide a viable non-parametric alternative to nonlinear parametric methods that often tend to overfit when only small datasets are available. The prediction model that relies on data for its forecast can further be integrated into the optimization framework for online correction in the dispatch problem based on the past prediction error.
Energy hubs are multi-carrier energy systems that incorporate conversion, storage and network technologies of different energy carriers such as electrical, thermal and/or chemical processes that follow individual demand and supply patterns under distinct time scales. Energy hubs facilitate easier integration of intermittent renewable energy sources, greater flexibility and higher efficiency of the overall system. Furthermore, the joint operational optimization of multiple hubs in the network that considers the energy trading between them further increases flexibility and RES utility, and reduces the overall operational cost of the system. The resulting system is also better equipped to meet challenges such as demand, generation and price uncertainty.
A plug and play framework that models a network of energy hubs has been designed along with a novel distributed control strategy for the multi-objective optimization of the network. The cooperative operation of the network and control strategies rely on future demand profiles for the energy and heating demand of the hubs based on the model of the hub. The ability to predict a hubs energy consumption, using a variety of performance metrics is crucial for future energy scenario planning. Data Driven modelling approaches are increasingly utilized since first principles models may be costly to develop, difficult to maintain and in some cases, first principle identification methods cannot be effectively employed. The increased quantity and quality of data collected facilitates the utilization of data-driven approaches, thereby realizing the potential for energy prediction as a complementary or alternative option to the more traditional physics based approaches and reducing the reliance on first principle models. These methods can also be tailored to factor in the seasonal and price dependency of a load. Data driven methods such as Gaussian process regression are also flexible due to its inherent ability to describe uncertainty in the prediction making it a suitable candidate for stochastic optimization and provide a viable non-parametric alternative to nonlinear parametric methods that often tend to overfit when only small datasets are available. The prediction model that relies on data for its forecast can further be integrated into the optimization framework for online correction in the dispatch problem based on the past prediction error.
The goal of this project is to design and evaluate different data driven techniques to model demand patterns of an energy hub. The developed method should account for the dependence of a load on external factors such as temperature, price, etc. as well as which of the loads are critical/fixed and which are deferrable. An additional aspect would be to investigate how the developed technique can be transferred and adapted to different hubs. Suitable methods should be identified, implemented and where necessary, adapted. The accuracy and results obtained using different models should be tested in simulation, as well as tested and cross-validated on real historical measurement data of the NEST demonstrator at Empa. Finally, the demand prediction method shall be incorporated into an existing MPC framework.
This is a research oriented project and a learning opportunity for both sides. We will learn about theoretical and practical aspects of energy demand forecasting, modelling techniques, using algorithms in machine learning and system identification, and concepts in nonlinear dynamics and power systems. The project can also be modified based on the interests of the student.
The goal of this project is to design and evaluate different data driven techniques to model demand patterns of an energy hub. The developed method should account for the dependence of a load on external factors such as temperature, price, etc. as well as which of the loads are critical/fixed and which are deferrable. An additional aspect would be to investigate how the developed technique can be transferred and adapted to different hubs. Suitable methods should be identified, implemented and where necessary, adapted. The accuracy and results obtained using different models should be tested in simulation, as well as tested and cross-validated on real historical measurement data of the NEST demonstrator at Empa. Finally, the demand prediction method shall be incorporated into an existing MPC framework.
This is a research oriented project and a learning opportunity for both sides. We will learn about theoretical and practical aspects of energy demand forecasting, modelling techniques, using algorithms in machine learning and system identification, and concepts in nonlinear dynamics and power systems. The project can also be modified based on the interests of the student.
Please send an email to:
Varsha N. Behrunani
bvarsha@ethz.ch
Please send an email to: Varsha N. Behrunani bvarsha@ethz.ch