The project used a high degree of modeling to support both predictive feeder response as well as smoothing and shifting algorithm development. The modeling effort used feeder data, predictions of PV energy output, battery response and response of the micro-grid in both EPRI’s Open Distribution System Simulator (OpenDSS) as well as Pacific Northwest National Laboratory’s GridLAB-D software. The battery-response modeling allowed development and optimization of both smoothing and shifting algorithms.

Cloudy PV production

The smoothing algorithm, developed by Sandia National Laboratories, resides within the battery controller. This power-based functionality requires fast reaction times to smooth intermittent PV output effectively, with the PV kilowatt output acting as a control signal. Because batteries are still a high-cost material, optimizing the degree of smoothing needed versus battery energy use is a critical endeavor. The algorithm has two modes of operation, a low-pass filter and a moving average. The gains, time constants and moving average window all are adjustable from the back-office software to ensure optimization. Both the moving average and low-pass filter were evaluated for effectiveness as a function of optimized battery use. Additional functionality was embedded into the smoothing algorithm by providing two auxiliary inputs to the logic. These auxiliary inputs were used for the coordinated control of the NEDO micro-grid.

The shifting control was designed using an algorithm residing in the PNM back office. The shifting algorithm is hosted in the back office because of the data types needed and the desire not to transmit extensive and sensitive data to field devices, conserving communication bandwidth and data security. Because of the lack of a commercially available grid-level battery control system, the shifting control strategy was based on calculating a setpoint for the system using an advanced calculation engine that sends the command to the battery control system. The algorithm polls real-time power market prices, supervisory control and data acquisition (SCADA) data and weather predictions from the National Oceanic and Atmospheric Administration (NOAA).

Based on these and other parameters, the shifting algorithm optimizes the delivery of stored energy on a day-to-day basis. To shave peak, the algorithm predicts the next day’s peak feeder load by correlating weather data, historic feeder loads, PV-output prediction, and correlations of different intermittences and types of cloud cover.

The building micro-grid operated by NEDO uses multiple variables to optimize building performance through the micro-EMS. Specific tests of coordinated control between the micro-grid and PNM’s battery storage system have recently been performed to study how the two systems can collaborate to benefit the local distribution system.

Utilities will have an increasing need to understand and collaborate with customer-owned distributed resources for overall system balancing. In this demonstration, data from the PV and battery system are actively passed every second to the building micro-grid, while the building micro-grid passes 1-sec data from the NEDO gas engine and fuel cell to the PNM battery control system. PV intermittency data is used as an input to the natural gas generator and fuel cell to allow the two resources to provide ramping support. Simultaneously, data is passed from the natural gas generator and fuel cell to the battery control system. The smoothing algorithm uses this data to reduce the duty on the battery system.

Demonstration Results

Peak shaving

The developed algorithms have successfully addressed both smoothing and simultaneous energy storage for dispatch at a time more valuable to the utility. Through simultaneous operation, the algorithms are able to take extremely variable PV output that typically would be difficult to regulate and create a defined output in both amplitude and duration during the system peak. The storage system is able to take available energy produced by PV, store it in the battery system with very little production to the grid and dispatch it as a predictable output, providing a firm block of energy aligning with the evening peak load.

The peak-shaving algorithm has been optimized and refined, and has enabled PNM to meet a target 15% reduction in peak summer load on the associated feeder. The shifting control algorithm understands the need for a peak load reduction on the feeder as opposed to a firming control. The algorithm dispatches the battery system to successfully reduce the peak that the distribution system would otherwise have to manage.

Coordination between the building micro-grid and battery storage system was demonstrated. Multiple models and simulations were conducted using archived data from the individual systems. The models studied optimization of the resources and created setpoints for the smoothing algorithm variables such as gains and time constants. Data from the NEDO micro-grid and energy storage site were exchanged on a cloudy day in a closed-loop control system. Results of the interaction are currently under analysis.

Next, during a clear day, PV intermittency was simulated by opening a breaker to disconnect a subset of PV strings. This controlled disconnection provided a known drop-in PV output that could be used more precisely for calculating response by the battery system as opposed to a variation based on the degree and opacity of cloud cover.