Method of operating a power distribution system

Patent metadata

Field	Value
Publication number	US-12040614-B2
Application number	US-201917254549-A
Country	US
Kind code	B2
Filing date	Jun 20, 2019
Priority date	Jun 26, 2018
Publication date	Jul 16, 2024
Grant date	Jul 16, 2024

Abstract

The invention relates to a method of operating a node of a power distribution system in an optimised manner so as to minimise transmission losses on an external power bus. A power generator feeds in electrical energy via an inverter capable of carrying out Volt/VAR control under the command of an optimiser associated with a supervised learning model such as a support vector machine.This optimisation takes place in two distinct training phases, one in which the supervised learning model is trained, another in which optimal control parameters for the inverter are obtained.

First claim

The invention claimed is:

1. Method of operating at least one node of a power distribution system comprising a power distribution bus, said node comprising:
at least one local power generator connected to said power distribution bus at a respective connection point via a respective inverter, said inverter having a power factor able to be predetermined by means of a controller and having a voltage/reactive power transfer function determined by said controller on the basis of a plurality of control parameters;
a plurality of sensors arranged to provide timestamped measurements of at least the voltage, active power and reactive power at said connection point;
an emulator adapted to emulate the behaviour of said controller and said inverter so as to output a value of emulated reactive power on the basis of an input value of said voltage, said emulator emulating said transfer function;
a supervised learning model; and
an optimiser arranged to output control parameters for said controller and to determine their optimum values;

wherein said method comprises a first training phase in which:
a first training dataset is obtained, comprising timestamped measurements of said voltage, said active power and said reactive power at said connection point obtained at a plurality of time under a predetermined initialisation set of control parameters provided to said controller;
said supervised learning model is trained on the basis of said first training dataset so as to determine a relationship between, on the one hand, measured reactive power and measured active power, and on the other hand, measured voltage at said connection point;

and a second training phase, in which:
a second training dataset is obtained, comprising further measurements of said voltage, said active power and said reactive power at said connection point obtained at a plurality of timestamps under non-optimised control parameters;
said further measurements of voltage are input to said emulator so as to generate an emulated reactive power value on the basis of said predefined control parameters supplied by said optimiser, a sum of said emulated reactive power value and said measurements of reactive power being input as a first input to said supervised learning model, said measured active power being input as a second input to said supervised learning model, this latter outputting, on the basis of its said inputs, a simulated voltage value which is input to said optimiser together with said sum of said modelled reactive power and said measurements of reactive power, said optimiser applying a gradient-free optimisation so as to iteratively determine optimised control parameters;
and wherein, in an optimised operating mode, said optimised control parameters determined in said second training phase are fed to said controller so as to determine the transfer function of said inverter on the basis thereof, and thereby to convert power from said local power generator to AC power at a power factor determined by said controller on the basis of said transfer function and the measured voltage at the connection point.

2. Method according to claim 1 , wherein, in said first training phase, said measurements of said voltage, said active power and said reactive power at said connection point are obtained at a plurality of times with the power factor of said inverter set to a value of 1.

3. Method according claim 1 , wherein, in said first training phase, said measurements of measurements of said voltage, said active power and said reactive power at said connection point are obtained at a plurality of times with said transfer function of said inverter set arbitrarily.

4. Method according to claim 1 , wherein said optimiser applies a Nelder-Mead method, or a Cobyla optimisation, or a genetic algorithm, or a particle swarm optimisation, or a Powell optimisation.

5. Method according to claim 1 , wherein said optimised control parameters are determined for a plurality of said inverters simultaneously.

6. Method according to claim 1 , wherein further measurements are taken during said optimised operating mode and are used to further optimise said optimised control parameters.

7. Method according to claim 1 , wherein said optimised control parameters are determined in said second training phase by minimising a cost function, said cost function being:

min
⁢

C
⁡
(

x
,
λ

)

=

λ
⁢

V
2

(
x
)

+

(

1
-
λ

)

⁢

L
2

(
x
)

V
⁡
(
x
)

+

L
⁡
(
x
)

where x is a value evaluated during the optimization process, and λ is a weighting factor between 0 and 1, V is defined as:

V
⁡
(
x
)

=

∑

n
=
1

N

⁢

1
K

⁢

∑

k
=
1

K

⁢

(

10
⁢

u

k
,
n

(
x
)

Assignees

• Csem Centre Suisse Delectronique Et De Microtechnique Sa—Recherche Et Dev
• Csem Ct Suisse Delectronique Microtechnique Sa Rech Developpement

Inventors

• Martin William

Classifications

• H02J2103/30

  Simulating, planning, modelling, reliability check or computer assisted design [CAD] of electric power networks · CPC title

• H02J2101/28

  Wind energy · CPC title

• H02J2101/24

  Photovoltaics · CPC title

• H02J3/50

  Controlling the sharing of reactive power · CPC title

• H02J3/48

  Controlling the sharing of active power · CPC title