Composing Modeling and Simulation with Machine Learning in Julia (2105.05946v1)

Abstract: In this paper we introduce JuliaSim, a high-performance programming environment designed to blend traditional modeling and simulation with machine learning. JuliaSim can build accelerated surrogates from component-based models, such as those conforming to the FMI standard, using continuous-time echo state networks (CTESN). The foundation of this environment, ModelingToolkit.jl, is an acausal modeling language which can compose the trained surrogates as components within its staged compilation process. As a complementary factor we present the JuliaSim model library, a standard library with differential-algebraic equations and pre-trained surrogates, which can be composed using the modeling system for design, optimization, and control. We demonstrate the effectiveness of the surrogate-accelerated modeling and simulation approach on HVAC dynamics by showing that the CTESN surrogates accurately capture the dynamics of a HVAC cycle at less than 4\% error while accelerating its simulation by 340x. We illustrate the use of surrogate acceleration in the design process via global optimization of simulation parameters using the embedded surrogate, yielding a speedup of two orders of magnitude to find the optimum. We showcase the surrogate deployed in a co-simulation loop, as a drop-in replacement for one of the coupled FMUs, allowing engineers to effectively explore the design space of a coupled system. Together this demonstrates a workflow for automating the integration of machine learning techniques into traditional modeling and simulation processes.

• The paper demonstrates JuliaSim's integration of traditional simulation with ML using CTESNs to accelerate complex engineering models.
• It leverages pre-trained surrogates and ModelingToolkit.jl for flexible transformations and rapid deployment in simulation tasks.
• Numerical results in HVAC applications show a 340x speedup and less than 4% error, underscoring its efficiency in design optimization.

Overview of JuliaSim: Integrating Modeling, Simulation, and Machine Learning

The paper presents JuliaSim, a high-performance programming environment designed to facilitate the integration of traditional modeling and simulation with ML. This integration aims to address the computational expense associated with detailed multi-physics component models, which constrains design, optimization, and control processes. JuliaSim offers a solution by enabling the creation of accelerated surrogate models leveraging continuous-time echo state networks (CTESN) within its framework.

Central Components of JuliaSim

JuliaSim is underpinned by ModelingToolkit.jl, an acausal modeling language. This enables users to compose and transform trained surrogates in its compilation process. The modeling environment accommodates both exact and inexact transformations, providing flexibility in leveraging AI/ML techniques alongside traditional symbolic transformations. Central to this approach is the utilization of CTESNs, which are adept at handling the stiff equations that commonly arise in engineering simulations.

The JuliaSim model library complements this by offering a collection of differential-algebraic equations (DAEs) and pre-trained surrogates. This library allows users to bypass the training cost of surrogates by utilizing pre-trained models, thereby facilitating their integration into design and optimization processes.

Applications and Numerical Results

The efficacy of JuliaSim and its surrogate acceleration is illustrated through its application to Heating, Ventilation, and Air Conditioning (HVAC) dynamics. The paper reports that the CTESN surrogates can accurately model HVAC cycles with less than 4% error while providing a simulation speedup by a factor of 340. In global optimization of simulation parameters, this approach yields a speedup of two orders of magnitude in identifying optimal settings.

Additionally, JuliaSim facilitates seamless co-simulation with external models via the Functional Mock-up Interface (FMI) standard. This interoperability allows engineers to explore design spaces effectively by deploying the surrogate as a drop-in replacement for coupled Functional Mock-up Units (FMUs).

Implications and Future Prospects

The integration of machine learning into modeling and simulation environments like JuliaSim has significant implications. Practically, it enhances computational efficiency, allowing complex models to be simulated and optimized more rapidly, which is particularly advantageous in fields requiring real-time control and immediate feedback.

Theoretically, this research demonstrates the potential for generalized surrogate models to provide robust approximations across diverse scenarios, thereby laying the groundwork for further advancements in automated model reduction techniques. Future developments may include embedding surrogates as FMUs for broader platform integration and exploring additional surrogate modeling techniques.

In summary, JuliaSim exemplifies a sophisticated approach to integrating machine learning with traditional engineering simulations, thus paving the way for more cost-effective and computationally efficient modeling in various applied domains. The methodology and results outlined in this paper will likely stimulate further innovations in the landscape of simulation and modeling technologies, expanding their applicability and utility across multiple disciplines.