Renewable energy error compensable forcasting method using battery
View Patent ↗A renewable energy error compensable forecasting method using a battery is provided. The method may include modeling a policy π θ with a parameter θ, by using a reinforcement learning algorithm in which a state s t of an agent in a time zone t and, a reward r t+1 in a time zone t+1 and a next state s t+1 with respect to an action α t taken by the agent are determined; determining the parameter θ capable of minimizing a sum of an error function f t + 1 D ; and generating a compensable error with the battery using a model including the parameter θ.
1 . A renewable energy error compensable forecasting method using a battery comprising:
modeling a policy π θ with a parameter θ, by using a reinforcement learning algorithm in which a state s t of an agent in a time zone, and, a reward r t+1 in a time zone t+1 and a next state s t+1 with respect to an action a t taken by the agent are determined;
determining the parameter θ for minimizing a sum of an error function
f
t
+
1
D
;
ana
generating a compensable error with the battery using a model including the parameter θ; and
charging or discharging the battery based on the compensable error.
2 . The renewable energy error compensable forecasting method of claim 1 , wherein the charging or discharging the battery based on the compensable error comprises charging the battery by the compensable error in case of under-forecasting.
3 . The renewable energy error compensable forecasting method of claim 1 , wherein the charging or discharging the battery based on the compensable error comprises discharging the battery by the compensable error in case of over-forecasting.
4 . The renewable energy error compensable forecasting method of claim 1 , wherein:
the state s: is determined using Equation 1 below,
s
t
=
(
o
0
,
o
1
,
…
,
o
t
)
[
Equation
1
]
(here, o t denotes an observed value in the time zone t).
5 . The renewable energy error compensable forecasting method of claim 4 , wherein:
the observed value O t is determined using Equation 2 below,
o
t
=
(
x
t
,
E
t
)
[
Equation
2
]
(here, x t denotes a measured value of renewable energy generation amount in the time zone t, and E t denotes energy stored in the battery).
6 . The renewable energy error compensable forecasting method of claim 1 , wherein:
the reward r t+1 is determined using Equation 3 below,
r
t
+
1
=
-
f
t
+
1
D
[
Equation
3
]
(here,
f
t
+
1
D
denotes an error function).
7 . The renewable energy error compensable forecasting method of claim 1 , wherein:
an objective function of the reinforcement learning algorithm is determined using Equation 4 below,
[
Equation
4
]
minimize
{
a
t
}
t
=
0
∞
𝔼
{
x
t
+
1
}
t
=
0
∞
[
∑
t
=
0
∞
γ
t
f
t
+
1
D
]
=
maximize
θ
𝔼
{
x
t
+
1
,
a
t
}
t
=
0
∞
[
∑
t
=
0
∞
γ
t
r
t
+
1
]
(here,
{
a
t
}
t
=
0
∞
denotes a forecasted value for minimizing the sum of the error function
f
t
+
1
D
,
γ t denotes a depreciation (0<γ<1), and
{
x
t
+
1
}
t
=
0
∞
denotes a future renewable energy generation amount).
8 . The renewable energy error compensable forecasting method of claim 7 , wherein:
the action a t is extracted π θ (·|s t ) corresponding to the policy in the state s t .
9 . The renewable energy error compensable forecasting method of claim 1 , wherein:
the reinforcement learning algorithm is implemented using an artificial neural network model.
10 . The renewable energy error compensable forecasting method of claim 9 , wherein:
the artificial neural network model updates the parameter θ by using a value function V θ (s t ) estimated from the state s t of the agent in the time zone t.