Using contextual information to increase the power of analyses of energy performance

Submitted by Bruno Santos Pimentel on Fri, 09/06/2019 - 17:31

Understand how to combine information about energy consumption, production, and environmental (or contextual) variables to improve analyses and decisions regarding energy performance


Imagine the following scenario. You are responsible for a critical process of your industrial plant, and you would like to better understand how to improve the energy performance, whether by investing in operational processes, investing in improvements in the equipment itself, or a combination of both.

So you decide to collect data about this process, and you soon find a recent record of consumption of the principal energy input of that process (in this example, we assume that that it is argon gas, and the consumption is measured in Nm3):

graphic 1


Consumption appears to decrease slightly during the period, but the data still don't help much in the analysis of energy performance. But you know that the levels of production of this process can vary, and of course this influences the consumption of argon:

graphic 2


“Wow,” you think, while reflecting on the graph. There does not appear to be a clear relationship between the two variables... Then you remember some concepts of statistics and see that the correlation between the two samples (consumption and production) is in fact very low (ρ=0.00159). That is, while the correlation does not imply causality, there does not appear to be a significant linear relationship between the two variables – which complicates the analysis somewhat, since the existence of a linear relationship would be the simplest assumption that could be made about the process at that moment.

“Oh, but what about the specific consumption?” you remember. In fact, the consumption of the energy input for each production unit (in our example, measured in (Nm3/t) should also be “well-behaved” – the higher the production, the lower the specific consumption, which should asymptotically approach zero.

graphic 3


“Now I see,” you say, (partly) relieved. Even with a small variation, the specific consumption does in fact appear to be influenced by production. “If it were possible to work with higher levels of production, the process would have better energy performance,” you conclude correctly – of course, up to a certain point.

This conclusion helps us in the analysis, but is not sufficient for understanding the behavior of the process to the point of being able to infer actions to improve its performance.

“Well - but there's just one thing: the process works with a mix of products that vary during the period...,” and it's very likely that each product has a different energy demand according to its physical characteristics and its interaction with the equipment. Fortunately you also have the consumption and production data for each of the four products of the mix (represented here by their codes “1,” “15,” “18,” and “22”).

graphic 4


At this point, you start to get excited about the results. The graphic begins to make more sense, and the energy behavior of the process seems to be more consistent with your experience and with the physical and operational fundamentals.

“There really is a linear relationship between consumption and production.” Yes, but it was masked by the average of the aggregated consumption and production of all of the products that are processed by the equipment in the analysis. By decomposing the original data into the respective consumption and production values of each product, the relationships become more evident.

You again make use of a simple but powerful statistical tool: linear regression. Using linear regression, the series of data of each product can be approximated by linear functions, making clear the fixed quantities (intercept) and variable quantities (slope) of energy consumption./

graphic 5

The linear equations that describe the relationship between consumption and production of each of the products are apparently well behaved, with good coefficients of determination. “But how can I use this model to drive actions for improving energy performance?”

The magnitude of the intercept is generally related to the volume of energy that is consumed regardless of production. In batch processes, for example, this fixed consumption can represent the energy that is consumed in the intervals of each run (or round). In continuous processes, on the other hand, the intercept can represent the basal energy consumption of a process. In both cases, identifying means for reducing fixed consumption implies direct gains for the efficiency of the process – reducing the power of ventilators of dust removal systems in the intervals between batches and improving the thermal insulation of ovens are examples of possible actions in the direction of increasing energy performance.

The slope of the line, on the other hand, is directly related to the ability of the process to convert energy inputs into production results. The greater the slope, the more energy intensive the process is, and, of course, the more important it is to identify the means to reduce its energy consumption. Improving the efficiency of burners in furnaces and boilers is an example of actions in this vein.

And of course, as you have already noted, increasing the levels of production – that is, reducing as much as possible the periods in which there is energy consumption without corresponding production – has a direct positive impact on the specific energy consumption of the process. The plots below make this quite clear, even though you know that there are limits to how much operational efficiency can be achieved in the process.

graphic 6

Finally, you can breathe (a little). “Any analysis becomes richer with contextual information. And this is even more important for processes with higher levels of variation in the parameters of interest.” In fact, the analysis of consumption and production data, now with information about the operational context (in this example in particular the mix of products), makes it possible to achieve significant insights about possible actions for improving the energy efficiency of industrial processes – something that would have been impossible using only averaged information originally considered for the analysis of the process.

“Understanding the production process well, the characteristics of the physical assets involved, the products, the energy inputs, and the operational methods applied is fundamental for any energy efficiency initiative,” you happily conclude, and head out for lunch.

Also read: How to build energy consumption prediction model

Product Manager, Viridis

Viridis Product Manager, with more than 20 years of leadership in innovation and technology programs in large industrial organizations. PhD and masters degree in computer science from UFMG, bachelor’s degree in mechanical engineering, innovation and sustainability fellow at Sloan School of Management, MIT. Extensive experience in project management and open innovation teams with industry, academia, and startups, applying digital technologies and analytics to challenges in productivity, strategy, and sustainable development.

Add new comment