AstraZeneca: Energy and Environmental Savings through Continuous Monitoring and Advanced Energy Analytics

The data consisted of approximately 8,000 readings of various parameters including temperature, pressure, and system status. To accommodate the size of the data set, additional servers had to be purchased to increase the system’s data storage and processing power. To manage storage capacity, data was captured in a 5 to 10-minute resolution and sent to the cloud servers for analysis twice an hour for access by the vendor. To automate this process, the system was programmed to schedule the data to be pulled from the building management system, configure the data points to a template recognizable by the analysis software, and time-sync a data dump from the local system to the cloud server.

The second step was to analyze the collected data using the third-party vendor’s data analytics algorithms. The results were instantaneously generated and displayed to the facility team via a web-based dashboard (see figure 1 for a screenshot). The dashboard includes a virtual sub-metering model built for the facility using the data from the Building Management System and equipment nameplate information. Any discrepancy to normal equipment operations that the algorithm finds is immediately reported via an email ticketing system. This ticketing system allows the vendor to identify an equipment parameter that is out of range and inform facility personnel. AstraZeneca can then immediately respond to make process changes to restore the system to normal operations and achieve optimal efficiency.

Figure 1: Virtual sub-metered energy consumption from an air handler unit in building 950, from site dashboard

Due to the relationship with the utility company for acquiring rebates and the effectiveness of the dashboard to verify savings, a 3-year monitoring contract with the vendor has been put in place. The cost includes energy analysis and dashboarding services. Additionally, a percentage of the utility rebate dollars from all identified controls projects are shared with the vendor. Half of the rebate funds to the site are dedicated to fund additional sustainability efforts. With the implementation of the continuous monitoring strategy, corrections to equipment operations can be done almost instantaneously. System errors have been corrected since the system came online, including settings that were inadvertently left in place that kept equipment running during unscheduled hours.

The optimal start control project came about from insights provided by the continuous data monitoring system. Before the optimal start controls were implemented, hardcoded schedules were utilized so the HVAC systems would start at designated times regardless of outside temperature or the building load. The use of the optimal start system saves 1.5 hours of system runtime most days by anticipating the heating or cooling needs of a workspace and starting the HVAC equipment at the right time to ensure optimal heating and cooling needs are met when the space is in use. The control system also utilizes outdoor air and considers building envelope conditions to determine the exact time each morning to begin heating and cooling.

No additional hardware was needed for the optimal start project since all configuration and testing were completed internally at AstraZeneca. The process of configuring the optimal start control was relatively simple. Once the settings were updated, the optimal start control was tested at a single building on just one Air Handling Unit (AHU) during Memorial Day weekend. The configuration parameters for the control tool had to be determined using a trial and error approach. Though it took almost 60 hours to get the first AHU running properly with the optimal start strategy, the same template was applied throughout the campus, which only required 1-2 hours per day to test each zone.  

Once the control strategy was implemented throughout the campus, facility maintenance personnel were trained on how the control system functions, but since the optimal start controls are self-learning this cuts down on engineering time and labor to determine appropriate start times and make seasonal changes manually. The algorithms make use of reinforcement learning techniques by correlating the operating conditions with past performances to identify the optimal controls for each AHU zone. During the learning phase, the algorithm builds a data table between outside air temperature and the time it takes to achieve the required room temperature. It took an average of 10-12 days for the algorithm to learn the optimal strategies for each system during the summer for cooling while the heating controls were leaned much faster. The total annual energy savings from the optimal start controls project was 732,465 kWh, 22,028 therms, which equals $96,435 per year in cost savings. See figure 2 for heat map of optimal start.

Figure 2: The heat map of the facilities energy consumption is plotted against the outside air temperature, showing energy reductions resulting from the optimal start strategy