19 / 44
Cognitive computing helps you outthink the weather
To convert weather data into useful load forecasts, data scientists
developed a comprehensive set of self-learning neural networks for
predicting load in different ISO zones. For each zone, more than 100
neural networks were trained using actual weather conditions.
Individual neural networks were trained to predict load for dif-
ferent types of days: regular weekdays, Saturdays, Sundays, and
holidays.
The load profile for each holiday is treated differently based on a
proprietary algorithmdeveloped by examining historical load profiles
on those days. This specialisation was further refined by training
multiple neural networks for bal-day, next-day and medium-range
forecasting for each day type.
Variable selection was used to optimise the appropriate set of
weather parameters needed for each zone, type of day, and forecast
period. The bal-day neural networks blend the most recent values of
observed load into the raw forecast values using a forward-correction
scheme similar to that used in FoD.
Make better decisions, with greater confidence
For utility companies, we see load forecasting as a critical key to
success. It is one that requires and deserves a sophisticated solution.
WSI Trader Load Forecast was designed so that energy decision-
makers can:
• Gain a competitive edge in both near-term and long-range time
periods by leveraging accurate, precise, and resolute data that
the company has available
• Distinguish between types of holidays and how they can impact
different sub-regions and zones by leveraging our proprietary
holiday-forecasting algorithms
• Get accurate bal-day and next-day forecasts based on our pro-
prietary forward-correction algorithms
• Leverage hyper-local forecasting designed to predict local weather
likely to affect load demand in the very near term
• Quickly visualise forecasts with the graphical, intuitiveWSI Trader
user interface
CONTROL SYSTEMS + AUTOMATION
Conclusion
In order to maintain consistent and reliable energy delivery − across
peak periods as well as everyday usage − decision-makers need to
leverage technically advanced load forecasting. Accurate weather
forecasting, combined with state-of-the-art data science, can poten-
tially help improve both short- and long-term forecasting.
Abbreviations/Acronyms
take note
Rob Berglund leads the sales team for Energy & Utility (E&U),
Agriculture, and Petro-Chemical for the Energy group at The
Weather Company. Since the beginning of his career at the
Weather Company, starting in 2003, he has developed business
in regions around the world, with an industry specific focus
of Energy & Water Utilities, Energy Commodities Trading,
Agribusiness, and Oil & Gas. Enquiries: Email
energysales@wsi.comThe two components of modern-day
weather forecasting are:
Intelligently using all available computer weather model
forecasts to provide the most accurate automated forecast
Having an expert and experienced local human forecaster
who knows the ‘local flavour’ of the weather and can
add further value (and better accuracy) than even the best
‘machine’ forecast
For South Africa specifically, the company employs the best weather
forecasting models, including the European Centre for Medium-Range
Weather Forecasting (ECMWF) model, the Global Forecasting System
(GFS) model and its proprietary high-resolution Deep Thunder model.
Given its high spatial resolution and advanced physics, the Deep
Thunder model is able to handle the localised weather features that
are unique to South African weather, from the daily ocean breezes
in coastal regions to the unique circulations associated with the
mountainous regions. Once the best forecasts are extracted from the
stream of the various weather forecasting models, an experienced
human forecaster is needed to improve the forecasts further. While
weather forecasting models are quite good, there are still flaws that
the expert forecaster can exploit, especially in extreme weather.
ECMWF – European Centre for Medium-Range Weather Forecasts
FoD
– Forecast on Demand
GFS
– Global Forecast System
ISO
– Independent System Operators
NAM
– North American Mesoscale
RPM
– Rapid Precision Mesoscale
THI
– Temperature-Humidity Index
WCI
– Wind Chill Index
• Weather can and does have a huge effect on performance
in utilities and industry.
• Predicting utility demand and consumption is a complex
and uncertain process.
• Accurate load forecasting depends on accurate weather
forecasting.
The Weather Company
17
February ‘17
Electricity+Control