Self-Learning Energy Management System on the Process Control Level

Abstract

An important part of increasing the energy and resource efficiency in companies is the reduction of energy consumption of production plants. In order to achieve this, suitable energy management concepts have to be developed. Energy management concepts involve collecting all required information and making decisions based on the evaluated data. This paper focuses on the approach of shutting down individual plant components in unproductive phases. Because manually shutting down and starting up plants is risky and time-consuming, plants are often left in a state in which they consume a lot of energy, despite not producing any parts, due to both scheduled and unexpected stops. For this reason, adequate energy management concepts are needed that automatically shut down unneeded plant components and restart them in time for the next productive phase. Multiple dependencies between plant components in the context of production and process flow lead to a massive increase in complexity. Subsequently, such concepts are rarely programmed in the control software. In this paper, we provide an approach that implements the energy management concepts as a superordinate entity at the process control level, which enables a holistic plant overview. Using flexible algorithms, the system should be able to make autonomous decisions about the ideal energetic state of the individual plant components. In order to minimize the effort of adding new plants, the developed algorithms should self-adapt to the respective plant configuration autonomously. In addition to machine learning algorithms, the functional analysis of production plants and knowledge concerning the structure of the plants gained from engineering tools are used.