Predicting peaks in power demand
Rising energy costs are accelerating the installation of energy management systems (EMS). But, once these systems are installed, what do you do with the data provided? Do these data show you where to save energy? Do they show how much energy you are wasting? While few things can replace the eye of a trained energy engineer, even the most skilled among them can use help in recognizing abnormal e...
Rising energy costs are accelerating the installation of energy management systems (EMS). But, once these systems are installed, what do you do with the data provided? Do these data show you where to save energy? Do they show how much energy you are wasting?
While few things can replace the eye of a trained energy engineer, even the most skilled among them can use help in recognizing abnormal energy usage. A well-designed EMS differentiates normal energy usage from abnormal. This is a difficult task because even normal energy consumption fluctuates from one facility to another, and can vary based on time of day, day of week, season of the year, ambient temperature, what process or operation is being run at that time or by the type of loads operating at any particular time. It would be very helpful if the EMS provided assistance in actually visualizing the energy consumption issues from this sea of data.
By comparing expected energy consumption patterns to present usage, deviations from modeled energy patterns can be used to flag areas of concern. These deviations point to targets as you hunt for energy consumption offenders in your facility. Although these deviations are time-stamped and tagged to a particular circuit, it is often very difficult to identify the issues if the data are presented in the usual report format. However, by applying pattern analysis techniques and using color and shape to convey the results, it is possible to provide users with a visualization that they can understand — even if they are not power experts.
Energy profiling tools for intelligent power meters that offer this type of visualization are now available. For example, a visual rendering of a circuit that operates similar processes during the same time on each day of the week is shown in Figure 1. Work performed at 10 a.m. on a Monday is similar to work performed at 10 a.m. on any other Monday, excluding holidays and shutdowns. Likewise, energy consumption patterns on Tuesdays should be similar to energy consumption patterns on any other Tuesday.
An energy profiling tool shows interval demand consumption through the color-coded stacked-bar chart with the height and color of the bar representing the magnitude of the demand for that interval. In this chart, each bar represents a 15-minute demand interval, with one 24-hour day holding a total of 96 intervals or bars. The background color used by this tool is based on the total energy consumed for that day. In this example, warmer colors have been chosen to indicate higher values while cooler colors are lower. Oct. 15 set a new monthly peak demand for that circuit, but is this considered abnormal energy consumption?
To help answer that question, the energy profiling tool has been told to compare demand time slices and total energy consumption for each day with the same day, but from a different week. When the tool was asked to examine each daily interval demand value and daily kWh value, a different color distribution was displayed in the report.
In Figure 2, the most severe anomalous energy consumption occurred, not on Oct. 15, but rather on Nov. 8 and Nov. 9. What is unusual for those days is that weekend energy consumption is normally very low. But in this case, energy consumption was at least 100% higher than the average for those time periods. The problem turned out to be workers coming in on the weekend and turning on equipment, but then leaving and not switching the equipment off. This equipment ran all weekend, causing the abnormal energy readings.
Other short term anomalous readings, shown in red, can be seen late in the day on Oct. 6, Oct. 8, Oct. 15, Oct. 28 and Nov. 18. Note that none of these instances set any new peak demands, but they were, in fact, incidents of wasted energy. A typical analysis of consumption patterns focusing on peak demand period would not have identified these areas of energy improvement opportunities.
Predicting energy consumption
An energy profiling tool can also predict what energy consumption is expected to be later in a day. This is done by comparing present consumption with historical averages and modifying the look-ahead predictions based on how closely the present consumption matches with the predicted values. If hour-by-hour energy consumption is higher than historical interval averages, the tool will appropriately increase its look-ahead prediction values giving you time to take action to reduce the possibility of new peak demands.
For example, recall from Figure 1 that a new peak demand was set on Oct. 15. This new peak turned out to be 9.2% higher than the previously recorded peak demand. For many rate structures, a peak demand charge can represent 50% or more of the total energy bill. In this case, the 9.2% higher demand raised the cost of electricity for that month by 4.6%. Even worse, some rate structures ratchet demand charges, resulting in even higher billings as the penalty continues for subsequent months. This billing increase was particularly unfortunate because it was detected three hours before the actual peak demand occurred (Fig. 3).
The lessons to be learned: It’s important to ensure that when alerts are received, they must be monitored, and that personnel are trained to know what to do when the system predicts that a new peak demand will be set. For suggestions on how to reduce demand if an alert is received, refer to “Choosing active solutions for power management” in the July 2009 issue of Plant Engineering magazine.
Fig. 1. Energy consumption is shown for 96 time slices of 15-minute increments each day with demand displayed as the vertical stacked bar chart and the kWh energy consumption for the day shown as background color for the day. The system normalizes the data so that times and days of high consumption can be easily identified. High energy and demand are shown as red and low as blue.
Fig. 2. An energy profiling tool can evaluate the energy consumption at a particular time and predict what it should be and compare it to the actual. Differences are highlighted and alerts are sent based on deviations from predicted to actual. A normal deviation chart is mostly dark stacked bars over a white background and indicates normal operation.
Fig. 3. At 10:47 a.m., the system alerted local personnel that a possible peak demand would occur later that afternoon. The new peak was, in fact, set at 1:44 p.m.