Forecasting & Modeling Design
RGA Labs specializes in both large and small scale custom built systems, using multiple models operating simultaneously. Our Metropolis™ architecture provides complete automation and data management in a fully modular environment. Rather than building a low priced "shrink wrapped" product, we custom tailor each system to your application. Although more expensive than off-the-shelf products, our custom systems exact the highest operating control performance and minimize the operations risk of your industrial processes. Failure of your process is the real cost of a modeling system. Metropolis™ provides the best forecasting and data management performance available at any price.
Our designs are particularly useful where extensive multi-systems interaction exist, and where the processes to be modeled are highly non-linear and/or non-stationary. We can modify and update your existing models and systems, or we can custom construct a complete package for your application. Applications include real time forecasting for utility and industrial processes, typically where weather has a strong impact on process variables.
- Energy Trading and Portfolio Management
- Real-time energy trading tools
- Electric Load Forecasting
- High-resolution 5 minute forecasting -- STLF ™
- Short term (168 hour) hourly forecasts -- STLF ™
- Medium term and seasonal demand forecasts
- Weather Data Management and Statistics
- Equipment Stress and Failure Prediction
- Electric distribution system outage forecasting
- Plant and power station equipment failure prediction
- Manufacturing Process Variable Control and Tuning
- Predictive Maintenance Programs
- High Performance Predictive Control Systems
Our forecasting and modeling systems are typically SQL database backboned. This allows for large scale asynchronous operation of multiple processes across multiple processors. Data and calculated results are fully time traceable allowing for continuous error performance monitoring. All input and output data is time tagged and stored for later analysis and recall. Historical conditions can be accurately recreated, and sequences of events can be faithfully recalled. This provides value both for continued modeling improvement as well as operational support.
Our model selection domain is extensive and allows us to bring the right tools
to the job. These model types include:
- Artificial Neural Networks
- Expert Systems
- Linear and Non-linear Regression Models
- Explicit Modeling
Metropolis™ can employ models both from RGA
Labs, as well as a large array
of different vendors, including your own internal models.
Since these modeling systems are all developed independently for each application, when deployed in a multiple model environment, time domain forecast uncertainties can be explicitly estimated.
Using our multiple model approach, your system's performance can be forecasted hours, days, weeks, months forward. The asynchronous multiple model strategy is deployed across the entire system allowing for:
- multiple forecast input streams,
- multiple forecast engines,
- a wide spectrum
of multiple output and data display strategies, and
explicit estimation of the forecast errors. - NOAA Port compliant systems
Because our systems are deployed on multiple processors and utilize fully independent
asynchronous processors, our forecasting systems are highly reliable. Failure
of a process or processor effects only a limited degradation of performance,
and overall system continues to operate. This is a particularly useful
feature when interacting with weather data sources, where the data environment
is very non-uniform.
In addition, this asynchronous strategy allows for real time, on-line development. Thus a system can be quickly installed with a minimum of features, and then expanded incrementally to meet your needs.
Most of our systems are deployed fully automated or with a high level of automation. This frees the user to apply their time resources towards interacting with the results and formulating new understandings about the processes being modeled. Data acquisition, data processing, model execution, database management, and results formulation and display are all executed autonomously. Process triggering can be accomplished through a variety of means including time clocks, interrupts from outside sources, and data conditions. This means that your process forecast is always current and available on demand.
RGA Labs partners with Tesla, Inc. for high performance regression technology. These models are well suited for time series forecasting of complex domains such electric load forecasting.
For more information on our forecasting technology, check out:
RGA Labs maintains full version control on all of its software and documentation.
We can provide you with a complete software library or maintain your library
in our shop.