Energy intensive industries had to find new ways of dealing with energy efficiency and resource management, due to the economic and financial crisis in recent years. Investments in energy efficient solutions and the replacement of old, inefficient technologies with new, more efficient, suggests a significant correlation between energy efficiency and the price of the final product. Furthermore, energy consumption is directly linked to the release of various emissions into the atmosphere, where exceeding limits represent additional costs. Hence, process optimization and management of energy and material resources is of crucial importance in such industries. Dynamic conditions on demanding international markets require a proper response of industrial companies to a variety of market signals and adjustments.
Conventional methods and tools for energy management, which are currently available, do not include the possibility of predicting the future process related states. Dynamic, short-term forecasts may enable early warnings and provide future insights into the short-term process states, while long-term forecasts enable decision support and scenario based response analysis, taking into account different market conditions. New methods of achieving savings in industry and the use of modern mathematical algorithms and artificial intelligence, represent an important step in the process of overall sustainable development of society. By combining advanced mathematical procedures with information and communication technologies (ICT), higher energy productivity of energy-intensive companies can be achieved.
In this dissertation modelling and contextualization of energy consumption in energy-intensive industries is presented. The main focus has been on energy consumption modelling and the context in which the representative consumption occurs. Furthermore, related resource, environmental and economy flows have also been assessed. At this point it has to be emphasised, that by taking into account a broader context of energy use, greater energy savings can be achieved in comparison to conventional process optimisation methods. The simulation algorithm of the proposed contextual model is based on neural networks. Production processes are described with energy cost centres (ECC), taking into account the identified specific contextual influential parameters.
The research confirms that measured historical data, enriched with context information can be used to identify energy profiles and energy consumption patterns in industry. For this purpose, two contextual models have been proposed, namely: a short-term context model for process optimisation and a long-term model for production planning and market response analysis. Production states used in the process of neural networks training have been collected through supervisory control and data acquisition system (SCADA) and implemented energy management system (EMS).
Development and application of models are shown in a case study of modelling the process of clinker burning in the cement production. To train the neural networks a Levenberg – Marquardt's back propagation algorithm has been used.
The proposed short-term contextual model has proved to be useful in the short-term forecasting of energy consumption and quality parameters of the final product. At this point it has to be emphasised that insufficient quality of the final product, results in production process prolongation. Short-term contextual model for quality parameters forecasts is based on NARX neural networks, trained with Levenberg – Marquardt's back propagation algorithm.
Developed long-term contextual model has been recognised as adequate to support long-term production planning in the cement production. By analysing different production outputs, fuel alternatives and related energy, emission and economical flows, context depended local optimums have been obtained. Contextual model for long-term production planning and decision support has been developed using Net Fitting neural networks, which have proved to be useful and sufficiently precise for long-term energy consumption prediction. Such forecasts are particularly important for scenario based analysis and cost optimisation, where the use of alternative fuels is highlighted. For long-term energy consumption forecasts a Levenberg – Marquardt's back propagation algorithm has been used.
The proposed methodology of contextual model of energy use in industry, offers functionalities beyond traditional industry solutions. Developed models combine basic principles of energy management systems, process control systems, decision support systems and modern mathematical approaches that have the power of forecasting.
Novel functionalities of the proposed contextual model and its robust structure represent an important upgrade to the traditional energy management systems. The use of contextual models can facilitate significant energy savings, assuming that target end users are properly trained and educated to make decisions related to energy efficiency. The possibility of wider integration of the proposed concept is clearly noticeable, especially in the scope of the transition of companies towards new industrial paradigm, Industry 4.0.