Siegmund Düll

Exploiting Similarity in System Identification Tasks with Recurrent Neural Networks

Abstract A novel dual-task learning approach based on recurrent neural networks with factored tensor components for system identification tasks is presented. The goal is to identify the dynamics of a system given few observations which are augmented by auxiliary data from a similar system. The problem is motivated by real-world use cases and a mathematical problem description is given. Further, our proposed model—the factored tensor recurrent neural network (FTRNN)—and two alternative models are introduced which are benchmarked on the cart-pole and mountain car simulations. We show that the FTRNN consistently and significantly outperformed the competing models in accuracy and data-efficiency.

Regularized Recurrent Neural Networks for Data Efficient Dual-Task Learning

We introduce a regularization technique to improve system identification for dual-task learning with recurrent neural networks. In particular, the method is introduced using the Factored Tensor Recurrent Neural Networks first presented in [1]. Our goal is to identify a dynamical system with few available observations by augmenting them with data from a sufficiently observed similar system. In our previous work, we discovered that the model accuracy degrades whenever little data of the system of interest is available. The presented regularization term in this work allows to significantly reduce the model error thereby improving the exploitation of knowledge of the well observed system. This scenario is crucial in many real world applications, where data efficiency plays an important role. We motivate the problem setting and our regularized dual-task learning approach by industrial use cases, e.g. gas or wind turbine modeling for optimization and monitoring. Then, we formalize the problem and describe our regularization term by which the learning objective of the Factored Tensor Recurrent Neural Network is extended. Finally, we demonstrate its effectiveness on the cart-pole and mountain car benchmarks.

Datenbasierte Optimalsteuerung mit neuronalen Netzen und dateneffizientem Reinforcement Learning

Zusammenfassung Methoden der Computational Intelligence werden im Automatisierungsumfeld zur Datenanalyse, Klassifikation, Regression, dynamischen Systemidentifikation, Zustandsschätzung, Steuerung oder Regelung eingesetzt. Dieser Beitrag zeigt, wie neuronale Netze, Regressionsbäume, Kernel-Methoden und Reinforcement Learning von der Datenanalyse bis zur fertigen Lösung eingesetzt werden können. Als Anwendungsbeispiel dient eine Gasturbine, für die optimale Regelgesetze aus Daten erlernt und bei Inbetriebnahme und Arbeitspunktoptimierung eingesetzt werden. Abstract