Motivation & TIM Liberalization of energy market in Europe in recent years have made balancing the electricity grid a challenging task. Nowadays, there are diverse consumption patterns and highly variable re-newable power sources like solar, wind, small-scale hydro, etc. at play. Energy companies need to forecast both consumption patterns of their clients and production capacities of their power sources in large numbers. TIM (Tangent Information Modeller) is a unified large-scale forecasting system written first in C++ and then in Julia. The engine builds time-series models automatically with no human intervention fulfilling the industry need for forecasting at scale. TIM ranked no. 1 in recent GEFCom 2017 competition (https://juliacomputing.com/press/2017/11/21/tangent-works-uses-julia-to-win-ieee-competition.html). TIM in Julia This high performance engine relies heavily on Julia’s computational paradigms like loop fusion, map, mapreduce, SIMD support, direct calls to BLAS etc.. TIM is AOT compiled and deployed in the cloud as a RabbitMQ worker. It is then used via an REST API in production by different energy platforms. In the talk, we would like to go through qualities of Julia which made this transition from C++ to Julia in production perfectly possible and share lessons learned. We also talk about entire architecture of the solution starting from the computational kernel all the way up to its REST API. Live demonstration showing how to set up your own forecasting system using TIM in Julia may also be included.
