CASTLE Lab has evolved a range of products and services. If you are interested in one of our more mature products, we can arrange projects through one of our consulting partners (or a consulting firm with whom you have an established relationship). We have worked directly with companies, through consulting firms or for investment banks to evaluate potential investment opportunities.
We also have a powerful library for modeling a wide range of dynamic resource management problems, which allows us to take on difficult new projects fairly quickly.
When appropriate, we will take on certain high-risk projects as research projects through Princeton University. All the software identified below is the property of Princeton University, and is licensed to companies through Princeton's Office of Research and Project Administration (ORPA). Do you think you have a problem that can challenge us? Contact Warren Powell, powell@princeton.edu, for information on working with CASTLE Lab.
Transportation and logistics
Activity forecasting
We have developed an operational forecasting system to forecast activities on daily (even hourly) basis, taking into consideration day of week effects, week of month, special days/holidays, and seasonal effects (e.g. the Christmas season). The system provides not just point forecasts, but also the entire probability distribution. Forecasting is performed simultaneously at multiple levels of aggregation.
Pilotview - Graphical diagnostics
Pilotview is an extremely flexible graphical display tool for displaying dynamic flows. It can show, using different modes of display, dynamic data for multilayer, multiattribute applications. For example, you may have a single layer (flow of a product), two layers (drivers and loads, locomotives and trains), three layers (pilots, aircraft and customers), or more. Each layer can be characterized by an arbitrary set of numerical and categorical attributes. Flows may be colored by categorical attribute.
A separate module within pilotview is designed to show the insights of a dynamic optimization model.
Custom modeling applications
We have addressed a broad range of problems that are best described under the broad umbrella of "dynamic resource management." We have a general purpose modeling library and solution strategy that has been applied to problems ranging from locomotives to the management of blood inventories to flu antivirals, R&D portfolio management to energy resource modeling. Applications range from strategic planning (pricing, analysis of operating policies, choice of equipment), short-term operational forecasting (e.g. for demand management), and real-time operations.
A central theme of our work is uncertainty. This provides several important capabilities:
- Risk analysis - We can assess the risk of potential policies and strategies. What is the likelihood that an asset will provide a return below some level? What is the probability that you will not be able to serve a customer request? What is the probability that you will exceed operational capacity?
- Robust solutions - Our methods can provide solutions that are robust, which is to say that they work well under a range of outcomes.
- Realism - For some applications, you simply cannot produce a realistic model without the ability to model uncertainty. For example, trains might be cancelled, customers might (or might not) arrive, equipment might fail. These discrete events cannot be handled using simple expectations (what do you do with 8/10 of a person?).