AI-ASSISTED ENERGY BIDDING FOR RENEWABLE POWER SOURCES

(1)

MANASSAKAN SANAYHA ([email protected]), PEERAPON VATEEKUL ([email protected])

DEPARTMENT OF COMPUTER ENGINEERING, FACULTY OF ENGINEERING, CHULALONGKORN UNIVERSITY, PATHUMWAN, BANGKOK, THAILAND

1

(2)

(3)

AI-ASSISTED ENERGY BIDDING

3

Decarbonization Decentralisation

Digitalisation Democratisation

(4)

¡ To propose a machine learning algorithm called “deep reinforcement learning” for

hourly day-ahead and intraday energy bidding in order to maximize the profit (lower the

costs).

¡ To investigate 2 proposed models supporting both market scales: wholesale and retail

energy markets.

(5)

5

1,885.96 kr. 1,910.13 kr. 1,932.33 kr. 1,937.85 kr.

1,646.02 kr.

551.44 € 484.01 € 637.25 € 527.39 € 420.41 €

0 500 1000 1500 2000 2500

Conv-A3C DPPO DDPG MBPG MB-A3C

Cost

Algorithm

A model comparison in terms of an average cost per day.

Average cost per day Denmark (DKK) Average cost per day Sweden (EUR)

For the Nord Pool scenario, MB-A3C provides the lowest average costs per day with 15% and 34%

reduction in Denmark and Sweden, respectively.

(6)

$790.51 $789.85

$738.10

$669.93

600620 640660 680700 720740 760780 800820

Energy Bill

A model comparison in terms of community’s energy bills ($)

For the Ausgrid scenario, MB-A3C3 is the winner

showing the most significant energy bill reduction for 15% (on 300 prosumers) compared with the energy

trading directly from grids. This is because we can sell energy at higher prices and buy it at cheaper prices.