While deep learning models provde robust performance in complicated tasks such as image segmentation and pattern recognition, a polynomial regression on the other hand is light weight and describable from its input to its output.
In a production plant, the energy consumption of the plant is roughly proportional to its machine operation and product output. Thus the regression model could rapidly adapts its weightage to the input and generates the prediction onsite. The model is extremely useful for the energy consumption estimation, failure detection, and incident recognition.
A flow diagram of chemical plant energy consumption model using big data is shown in figure below. The data is collected from machine sensors and output dash board. Historical data is used for the initialization of the weightage of the regression model.
Given that the historical data has been collected with labeling, the subsequent data processing and visualization should occupy large fraction of the time. Preliminary searching of the benchmarking model and the data pattern visualization are established quickly at the begining. The data and model are then refined within few rounds of iterations.
Correlations between features (usually the inputs) and output are established in a dash board. The critical features are retained in the model whereas unrelated or noise data are discarded.
Comments