Prediction of Compressed Air Demand Depending on the Type of Production with the Use of Neural Networks

Compressed air systems are commonly used in industrial plants to produce the compressed air required for the facility's daily operations. Since air compressors consume more electricity than any other type of facility equipment, an optimization of the efficiency of compressed air system operation cycles is essential for energy savings. In this article the demand for compressed air in production plants with different operating characteristics is analyzed. It is checked how the neural network identified for a given plant would work in the case of another plant with a different needs while predicting compressed air demand, which is understood as a prediction of compressor on/offs. The simulation results based on real data indicate possible decisions that improve system efficiency. LSTM network seems to be well suited for identification achieving best results on dedicated object used for training. Cooperation of neural network updated in real time with supervisory controller may achieve little margain error and provide accuarte control system decision support.