International Core Journal of Engineering 2020-26 | Page 27

IV. P OWER M ARKET S ETTLEMENT D ATA R ISK P ROCESSING i is settlement interval.
A. Analysis of Risk Processing of Power market Settlement
Data
In the process of settlement of the power market, data
risks always exist. After the abnormal data is identified by
the evidence theory, the identified abnormal data needs to be
processed. In the spot market, due to the short settlement
interval, more settlement data, how to process the identified
abnormal data quickly becomes an important part of the
settlement data risk. The paper will study the risk processing
of settlement data in combination with the actual situation in
the power market settlement and the above risk
identification results. After the anomaly data identification
of the data samples by the evidence theory, the degree of
support is obtained. Therefore, the paper proposes the
concept of reconstruction threshold T. If the reliability of a
certain data is higher than the reconstruction threshold T, but
lower than the lower limit of support, it means that the
evidence is abnormal data but its abnormality can accepted.
At this time, in order to improve the settlement efficiency,
the abnormal data may still be settled, but it needs to be
identified. After the settlement, analyze the cause of the
abnormal data and retrieve the correct settlement data for
error settlement. If the reliability of a certain data is lower
than the reconstruction threshold T, the deviation from the
normal level is excessive. At this time, if still settle by
abnormal data, there is a big risk. Therefore, it is necessary
to use a certain processing method to replace the abnormal
data with the reliability lower than T, so that reduce the
settlement risk. The polynomial curve fitting method only needs to
reconstruct the data near the data, which is easy to obtain.
However, when the settlement data fluctuates greatly, the
reconstructed data fitted by the polynomial will have a large
difference from the actual value. The historical correlation-
based method is applied to historical contemporaneous data,
which is more accurate when the settlement data fluctuates
greatly, but when the market entity operates for a short time,
it will lack relevant data.
According to the above analysis, the paper proposes a
combination of polynomial fitting and historical correlation.
The steps are described.
a) Perform polynomial curve fitting on the data
needed to be reconstructed;
b) Calculate the curvature of the fitted curve;
c) If the curvature is less than the threshold value Q,
the settlement data is considered to be less fluctuating, and
the fitting curve is used as the reconstruction data;
d) If the curvature is larger than the threshold value Q,
it is considered that the settlement data fluctuates greatly,
and the reconstructed data is calculated using the historical
contemporaneous data using Equation (21).
V. S IMULATION A NALYSIS
A. Power market Settlement Data and Processing
The paper selects a domestic power market as an
example. In addition, abnormal data needs to be set in the
settlement data. The paper selects each user to generate
abnormal data for a certain month. Taking User 1 as an
example, the settlement data is shown in Table I, in which
March 2018 is the generated abnormal data.
B. Power Market Settlement Data Reconstruction Method
Reconstruction methods for settlement anomaly data
generally have polynomial curve fitting [12] and
reconstruction methods based on historical correlation of
settlement data.
TABLE I. T ABLE F OR C ONSUMER ’ S S ETTLEMENT D ATA
1) Polynomial curve fitting reconstruction method: In
essence, polynomial curve fitting is a kind of linear
regression method, and its mathematical expression is
Equation (20).
y  x , w 
M
¦ Z x
j
(20)
j
j 0
Where M is the highest number of polynomials.
2) Reconstruction method based on historical
correlation of settlement data: The settlement price and
electricity are regularly followed, and closely related to time.
So historical contemporaneous data can be used instead
when resolving anomalous or missing data. In addition, the
paper considers the characteristics of settlement data over
time, and processes historical synchronic data through
Equation (21).
n
f  x 
f c  x   k
f c  x  
i 1
i 1
n
Settlement
price
/(¥/kWh) Market
electricity
/kWh Actual
electricity
/kWh
June 2017 0.21306 105000 105000
July 2017 0.18097 143000 143000
August 2017 0.12008 119084 119084
September 2017 0.1 160000 115098
October 2017 0.11699 140000 140000
November 2017 0.185 134000 117277
December 2017 0.22338 115000 113679
January 2018 0.232 110000 99120
February 2018 0.229 120000 115890
March 2018 0.12 178000 164980
April 2018 0.23937 81000 80360
May 2018 0.23655 99362 97227
The three-dimensional matrix X(I×J×K) will be
expanded in the direction of the settlement interval to obtain
X(20×36). PCA and ICA are required for the data samples.
When performing PCA, the percentage of each principal
component is calculated to determine the number of
variables that need to be retained after dimension reduction.
The paper selects the first 8 variables as the retained
principal components, and the retained information can
n
¦ f  x   ¦ f c  x 
Settlement
interval
(21)
Where f  x  is refactoring data, f c  x  is historical data,
5