IEEE Power & Energy Magazine - May/June 2018 - 58
Jiangsu province. the first is the demand-response potential
evaluation module; this module takes three steps to evaluate
the demand response potential of residents, commercial buildings, and industrial consumers. first, the basic electricity consumption patterns are extracted via clustering historical load
profiles. after that, the extracted load patterns among different consumers and different days are compared. finally, those
consumers with highly volatile load and low base electricity
consumption patterns are identified as high demand response
the second is a load forecasting module for large consumers. the module takes the basic consumption patterns of every
consumer from the historical load data to establish a Markov
model to describe the consumers' switching rule from one
pattern to another. finally, the module forecasts the next-day
load profile based on the Markov model. figure 4 shows the
clustering-based large-consumer load forecasting results from
Jiangsu province. compared with a support vector machine
(svM)-based method, the clustering-based method exhibits
superior performance, as indicated by the smaller distance
between the actual and predicted loads.
Simulation of Renewable Energy Outputs
because of renewable energy resources' production uncertainty, variability, and complex correlations, accommodating
high shares of renewable energy has been a major challenge
for the power industry. big data analytics can be used in the
analysis, forecasting, and simulation of renewable energy to
further increase its penetration. in particular, big data analytics has contributed to the simulation of renewable energy outputs and been embedded into the decision-making processes
for power system planning.
recently, china has seen a surge in the deployment of wind
turbines and solar photovoltaic (Pv) technologies. at the end
of 2016, the total installed capacities of wind and solar power
reached 169 gw and 77 gw, respectively. in addition to the
need for short-term renewable energy forecasting, the power
industry needs the chronological output series of wind power
and Pv, which is necessary for generation and network planning and determining future power system operation modes.
the industry cannot simply use historical data because
✔ the development of wind and solar generation is so
rapid that the accumulated historical wind power and
Pv data are not sufficient for the physical model to
simulate wind power and Pv outputs
✔ wind and solar power have been so heavily curtailed in
recent years that real-time measurements do not accurately reflect the potential for unconstrained existing
wind and solar power outputs, e.g., the measurements
usually fluctuate less than the full unconstrained potential output because output is curtailed.
therefore, the industry must rely on analytics to generate
such chronological output series.
several institutions have devoted a significant amount of
research efforts in wind and Pv power generation simulation.
ieee power & energy magazine
figure 5 shows the grid optimal Planning tool platform
developed by tsinghua university; (a) and (b) are the
basic user interfaces, with (c) and (d) the simulated wind
and Pv power outputs, respectively, of ten sites. these outputs have been widely applied to power system planning
the major challenge of simulating renewable energy
output is the consideration of the stochastic characteristics,
volatility characteristics, and temporal and spatial dependencies of wind speeds and solar radiations at different sites. it is
even more difficult to jointly consider all these aspects simultaneously. typically, a method that captures one characteristic well may not be able to incorporate the others.
to simulate wind power outputs, we first start with a windspeed profile generation. the core methodology is to use multidimensional stochastic differential equations (sDe) to generate
wind-speed time series, where the stochastic and intermittency
characteristic of each wind farm can be modeled by the drift
part of sDe and the temporal and spatial dependencies can be
captured by the multidimensional diffusion part in sDe. the
process is carried out as follows:
1) calculate the key characteristic indices of wind speeds
according to the historical wind-speed data, including the probability distribution, autocorrelation coefficient, cross-correlation coefficient, and monthly and
hourly average wind speeds.
2) apply the multidimensional sDe sets to generate a
multidimensional time series of wind speeds.
3) Modify the wind speeds considering the weak effect.
4) convert the wind speeds into wind power outputs using the power curve of a wind turbine.
5) sample the availability of each wind turbine.
6) obtain the simulated wind power outputs, based on
the wind power outputs considering the power curve
and the reliability model of the wind turbine.
compared with wind power output, Pv output has both
deterministic and stochastic parts. the deterministic part is
the solar irradiation of the sun and can be easily calculated for
each Pv station. the stochastic part is the shedding of production resulting from clouds and aerosols. the simulation of the
time series of the shedding effect is similar to that of wind
power. the process is carried out as follows:
1) use the global solar irradiation model to simulate the
solar irradiation intensity without shedding effects.
2) use the sDe to sample the clearness index that represents the effect of the shedding of clouds. calculate
the reflection and refraction coefficients according to
the clearness index and altitude of the site.
3) calculate the power output of Pv panels according
to the arrangement of these panels (inclination angle
and direction angle), tracking mode (fixed panel, horizontal single-axis tracking, tilted single-axis tracking,
and dual-axis tracking), solar irradiation, clearness index, reflection coefficient, refraction coefficient, and
the output characteristics of the Pv panels.