Smart Grid Seminar
Annenberg 213
Optimizing the Coordinated Control of Distributed Generation, Storage and Demand Response Based on Decision Trees
Panayiotis Moutis,
Electrical & Computer Engineering,
Carnegie Mellon University,
The questions regarding the provision of firm capacity by Distributed
Generation (DG) as also the coordinated control of its de-loading, so
that it can mitigate over-frequency phenomena, may be answered by
incorporating DG under the Virtual Power Plant (VPP) paradigm. A VPP
represents an abstract organizational concept of generation, storage
and demand units that can cater for the procurement of ancillary
services at Power System (PS) level, while ensuring optimality for the
units involved. Taking into account the stochastic nature of PS load
demand, as also of a great number of DG units based on Renewable
Energy Sources, the uncertainty implied requires for solutions that
(i) are characterized by flexibility, (ii) may not be optimal, but
represent the best available approach, and (iii) can be realized in a
short-time-ahead horizon, so that they exploit the most current
information.
To this end, binary decision trees (DT) based on the Shannon entropy
metric of information gain are used. Although the progress concerning
DTs has been vast and in-depth, there are features of the
aforementioned version of the tool that have not been clearly
identified and exploited, previously. Thus, a DT-based methodology is
suggested which can either re-dispatch the assets of a VPP, so that
they cover for a considerable loss of power and, thus, provide firm
capacity, or reduce the total output of the VPP, in order to support
the mitigation of some over-frequency in the PS. In both cases, the
optimality of the dispatching is ensured regardless of the stochastic
nature of the VPP components, while the hour-ahead horizon of the
realization exploits the current data. The time tedious process of
generating the learning set of the DT can be minimized by splitting
the burden among the microprocessors of the units of the VPP.
During the talk, the focus will be on (without being limited to) the
characteristics of the binary DTs as a tool which can approach an
optimization problem from the viewpoint of actual applicability;
especially, when emergency (contingency) measures are required. Based
on experience, a more generic discussion on the Smart Grid concepts
and their deployment in modern PSs may follow and is encouraged.
Additional info on the recent work and publications concerning the
talk can be found in http://panay1ot1s.com/
Generation (DG) as also the coordinated control of its de-loading, so
that it can mitigate over-frequency phenomena, may be answered by
incorporating DG under the Virtual Power Plant (VPP) paradigm. A VPP
represents an abstract organizational concept of generation, storage
and demand units that can cater for the procurement of ancillary
services at Power System (PS) level, while ensuring optimality for the
units involved. Taking into account the stochastic nature of PS load
demand, as also of a great number of DG units based on Renewable
Energy Sources, the uncertainty implied requires for solutions that
(i) are characterized by flexibility, (ii) may not be optimal, but
represent the best available approach, and (iii) can be realized in a
short-time-ahead horizon, so that they exploit the most current
information.
To this end, binary decision trees (DT) based on the Shannon entropy
metric of information gain are used. Although the progress concerning
DTs has been vast and in-depth, there are features of the
aforementioned version of the tool that have not been clearly
identified and exploited, previously. Thus, a DT-based methodology is
suggested which can either re-dispatch the assets of a VPP, so that
they cover for a considerable loss of power and, thus, provide firm
capacity, or reduce the total output of the VPP, in order to support
the mitigation of some over-frequency in the PS. In both cases, the
optimality of the dispatching is ensured regardless of the stochastic
nature of the VPP components, while the hour-ahead horizon of the
realization exploits the current data. The time tedious process of
generating the learning set of the DT can be minimized by splitting
the burden among the microprocessors of the units of the VPP.
During the talk, the focus will be on (without being limited to) the
characteristics of the binary DTs as a tool which can approach an
optimization problem from the viewpoint of actual applicability;
especially, when emergency (contingency) measures are required. Based
on experience, a more generic discussion on the Smart Grid concepts
and their deployment in modern PSs may follow and is encouraged.
Additional info on the recent work and publications concerning the
talk can be found in http://panay1ot1s.com/
For more information, please contact Sydney Garstang by email at sydney@caltech.edu.
Event Series
Smart Grid Seminar Series