Mohammad Gitizadeh

Mohammad Gitizadeh

Master's Thesis

Exploring the Potential of AI Solutions for Time Series Forecasting: Predicting Compressor Power Stages in Cooling Units of the Tucher Brewery Industry


Prof. Dr.-Ing. Thomas Seel, Simon Bachhuber (M. Sc.), Prof. Dr. Anne Koelewijn


01 / 2023 – 06 / 2023


Introduction to why this topic is so interesting and why there should be additional research, with references to current work [1, 2, 3, 4, 5, 6]:

Energy is a significant expense in the food and brewery industries. Food and brewery industries can achieve notable cost savings, maintain high product quality, meet regulatory requirements, and contribute to a more sustainable and environmentally friendly future by focusing on energy management and operation optimization.

One of the main energy consumers in the food and brewery industries is the cooling units which play a critical role in maintaining precise temperature and humidity conditions throughout various stages of production, storage, and distribution.

To gain effective energy management and practical operation optimization, we can focus on the heart of the cooling units: compressors. The purpose of the compressors in the cooling unit is to circulate the refrigerant (cooling agent) in the system under pressure to concentrate the heat it contains. By monitoring the behavior of the compressors, we can maintain the primary consumers of the energies in the cooling units and, at the same time, enhance the operation optimization.

To achieve this goal, food and brewery companies must develop an accurate model to predict compressor power stages over time. We would develop a project to forecast compressor behavior in response to the need and more detailed research in this area.

Artificial Intelligence (AI) based models have demonstrated remarkable performance in a wide range of tasks, including significant potential for predicting time series sequences. Deep learning models have a remarkable ability to learn arbitrary complex mappings from inputs to outputs and support multiple inputs and outputs sequences.

Due to deep learning models’ capabilities in predicting time series sequences, we have chosen deep learning-based solutions to predict compressor power stages in the Tucher brewery plant.

Explain the goals and required major steps to conduct the research in this thesis topic with references [4 ,5, 6, 7 ,8 ,9, 10]:

The goal of this Master’s thesis is to explore the potential of Artificial Intelligence (AI) solutions for predicting compressor power stages in the cooling units of the Tucher brewery plant. In order to maximize the prediction horizon, this thesis aims to examine and compare three different forecasting strategies: one-step prediction, multi-step prediction, and time series classification using various deep learning networks on the actual data from the Tucher brewery plant.


[1] Kaur, Devinder, et al. “Energy forecasting in smart grid systems: recent advancements in probabilistic deep learning.” IET Generation, Transmission & Distribution 16.22 (2022): 4461-4479

[2] Lin, Wen-Hui, et al. “Wind power forecasting with deep learning networks: Time-series forecasting.” Applied Sciences 11.21 (2021): 10335.

[3] Géron, Aurélien. Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow. ” O’Reilly Media, Inc.”, 2022.

[4] Brownlee, Jason. Deep learning for time series forecasting: predict the future with MLPs, CNNs and LSTMs in Python. Machine Learning Mastery, 2018.

[5] Khan, Anam-Nawaz, et al. “An ensemble energy consumption forecasting model based on spatial-temporal clustering analysis in residential buildings.” Energies 14.11 (2021): 3020

[6] Wang, Yunli, Chunsheng Yang, and Weiming Shen. “A deep learning approach for heating and cooling equipment monitoring.” 2019 IEEE 15th International Conference on Automation Science and Engineering (CASE). IEEE, 2019

[7] Gasparin, Alberto, Slobodan Lukovic, and Cesare Alippi. “Deep learning for time series forecasting: The electric load case.” CAAI Transactions on Intelligence Technology 7.1 (2022): 1-25

[8] Graves, Alex, Santiago Fernández, and Jürgen Schmidhuber. “Bidirectional LSTM networks for improved phoneme classification and recognition.” Artificial Neural Networks: Formal Models and Their Applications–ICANN 2005: 15th International Conference, Warsaw, Poland, September 11-15, 2005. Proceedings, Part II 15. Springer Berlin Heidelberg, 2005

[9] Yu, Yong, et al. “A review of recurrent neural networks: LSTM cells and network architectures.” Neural computation 31.7 (2019): 1235-1270

[10] Hewage, Pradeep, et al. “Temporal convolutional neural (TCN) network for an effective weather forecasting using time-series data from the local weather station.” Soft Computing 24 (2020): 16453-16482

[11] Kazemi, Seyed Mehran, et al. “Time2vec: Learning a vector representation of time.” arXiv preprint arXiv:1907.05321 (2019)