Buildings Participation in Resilience Enhancement of Community Microgrids: Synergy Between Microgrid and Building Management Systems

Military bases, communities, and university-scale microgrids are being implemented to serve critical loads of high priority. Remote switch scheduling and distributed energy resources (DERs) operation strategies for post-disaster service restoration have been explored in previous literature. Flexible buildings offer the central microgrid management system an opportunity to ensure available energy is directed to critical loads. This work presents a novel bi-level optimal sequence of operations for managing the controllable devices in small-scale microgrids to serve loads based on a priority scheme in campus-scale microgrids. This study introduces a technique for step-by-step restoration of customers’ granular loads. After and during an outage scenario, the facilities’ internal loads are energized in sequence according to their customer and local criticality levels, as well as the amount of energy available from the bulk system and DERs. The proposed methodology is formulated as a mixed-integer linear programming (MILP) model and adapts to various operating conditions. The proposed method is validated by performing controller hardware-in-loop (CHIL) case studies on the Banshee microgrid benchmark model on a real-time simulator.

INDEX TERMS Building management systems, demand side management, hardware-in-the-loop simulation, microgrids, power system restoration. 15

N n
The total number of nodes.    Power factor limit of an inverter. V min /V max ANSI standard min/max voltage limits. V 0 Nominal phase voltage. P 0 , Q 0 Nominal active and reactive power.
Priority index of the load L at node i.
Priority index of the load L with h th load at node i. 17

SET AND INDICES
t ∈ T Time sets. i ∈ N Node sets. (i, j) ∈ l Line sets. i ∈ L Load sets connected to node i. i ∈ g Fuel-based generation sets connected to node i. i ∈ v Intermittent generation sets connected to node i. i ∈ e Battery unit sets connected to node i h ∈ h Load group sets. The voltage at load L is located at node i at step t. 21 22 Every year the number of disaster events increases globally, 23 primarily attributed to climate change. Natural disasters can 24 be in several forms, such as landslides, hurricanes, tornados, 25 wildfires, etc. [1], [2]. In general, natural disasters can be 26 a potential threat to the electrical power system. However, 27 hurricanes are the major cause of damage and interruption 28 of electrical energy supply for long durations, which affect 29 economic prosperity, national security, public health, and 30 safety [3]. Most disaster outages result from power delivery 31 support structures damage and failure, which relates to distri-32 bution towers and poles [4]. Hence, the resilience of power 33 systems against hurricanes is critical to ensure their rapid 34 recovery. 35 Electrical power systems have been traditionally designed 36 to be reliable during typical contingencies. However, they 37 are now strictly required to be resilient against the high-38 impact and low-probable events caused by extreme weather 39 and cyber-physical events [5]. After a natural disaster, the 40 objective of the system operators is to restore the critical loads 41 as a priority and reduce the total number of de-energized 42 customers if it is not possible to restore all the loads. During 43 the outage period, consumers rely on backup systems to 44 maintain the power supply to the critical loads. Although, 45 most of these backup systems present a limited time/power 46 capacity and high operational cost. 47 The general concept of resiliency is the ability of a system 48 to anticipate and withstand external shocks, bounce back to its 49 pre-shock state as quickly as possible, and adapt to be better 50 prepared for future scenarios. These concepts become more 51 critical from the power systems scope under catastrophic 52 events [6], [7], [8]. Power systems resilience modeling must 53 consider the fragility of individual components and the whole 54 system interaction based on the lack of available resources. 55 Each device has a failure probability that relies on its histori-56 cal data and real-time usage level. 57 Critical consumers are not entirely composed of critical 58 loads but a mix of critical and non-critical devices. A grid-59 friendly demand response (DR) controller can efficiently 60 manage the building level loads based on their criticality 61 rank and the available power generation during the outage 62 period [9]. Advanced Building Energy Management Sys-63 tems (BEMS) technics allow a power consumption reduction 64 without compromising comfort [10], as well as support and 65 ensure the grid operation stability and safety as an ancillary 66 service [11], [12]. Residential facilities are compounded by 67 binary and flexible controllable loads, which both can be 68 optimally managed to achieve the facility's electrical goal 69 based on weather, grid, and consumer constraints [13], [14]. 70 Furthermore, once their communication with the distribution 71 network operator (DNO) is reliably and coordinately estab-72 lished, these distributed resources can strongly contribute to 73 grid resiliency [15], [16].

74
The exponentially increasing number of Distributed Gen-75 eration (DG) and Distributed Energy Resources (DERs) 76 VOLUME 10, 2022 integrated into the distribution systems have led to a decenliterature that works with the main grid and benefits both cus-126 tomers and DNO. In addition, numerous papers have also dis-127 cussed the possibility of BEMS solutions that can minimize 128 the customers' impact under DR periods, but there is still 129 a research gap on networked microgrids to meet resiliency 130 objectives.

131
There is a current lack of integrated MGMS and BEMS 132 solution for high-critically facilities that considers buildings' 133 flexibility for load shedding. Most research has approached 134 demand response without considering the physical limita-135 tions of real buildings. Critical facilities have a high level of 136 non-controllable loads, which are even more essential for site 137 operation.

139
The current study proposes a complete integrated MGMS and 140 BEMS solution. The novel operation strategy focuses on con-141 trolling DERs, flexible building loads, and remote switches 142 to maintain or drop the power supply to critical loads based 143 on a priority scheme. With a multi-time step optimization 144 technique in a hierarchical scheme, the strategy enables the 145 entire or partial restoration of the distribution system loads. 146 All levels work interactively in real-time to make decisions 147 based on measurements and grid conditions. The proposed 148 service restoration approach seeks to restore the system's 149 loads by evaluating the available power consumption range. 150 Upper control levels observe the network state to improve its 151 reliability and resiliency, as lower levels control optimized 152 loads and DERs to respect higher-hierarchical commands. 153 Each consumer can optimally manage which devices are 154 critical for its current operation with this range.

155
The main contribution of this work is to address the current 156 research gaps related to the lack of integrated microgrids and 157 building solutions to meet resiliency objectives.

158
The primary contribution of the paper is:

159
• The implementation of the granular load controls and 160 their integration with other agents inside the multiagent 161 platform.

162
• Networking building DERs allow for flexible coordi-163 nation among distributed devices for networked micro-164 grids. The proposed integrated solution will include 165 BMS using VOLTTRON to enable priority control of 166 loads.

167
• Networked buildings' load management will enable 168 serving more critical loads.

169
• Using several distributed control and sensing platforms, 170 VOLTTRON for integrating buildings will solve solves 171 communication bottlenecks.

174
• Interoperability issue mitigated by using several 175 building-level protocols/drivers supported in 176 VOLTTRON.

177
The remaining sections of this paper are organized as follows. 178 Section II clarifies the resilient-oriented building operations. 179 Section III presents some of the problem formulations for 180 emergency mode operation. Section IV describes the Ban-181 shee Microgrid testbed network. In section V, the case sce-182 narios are discussed, and the results are presented. Finally, 183 Section VI provides the conclusion of the present study.  [27], [28].

193
The novelty of this paper is the disintegrated control of  • Group 1: Electric vehicle charging equipment (5%).

247
Demand response approaches rely on reliable and stable 248 bi-directional communication between BEMS and electrical 249 sensors and controllers. The Internet of Things (IoT) has 250 been widely applied to energy systems like the Internet of 251 Energy (IoE), which encompasses the communication net-252 work responsible for linking central controllers and intelli-253 gent, flexible devices for data flow, commands, and cloud 254 storage [29]. With stable communication, building flexible 255 loads can be commanded by a high-level controller capable of 256 receiving and compiling information from a granular level to 257 define the best group scenario [30]. The granular level opera-258 tion is based on the assigned weighting factor of each device, 259 which is defined in each BEMS/HMS priority control algo-260 rithm and can vary during the operation period. Each building 261 has its composition of loads and the weighting factor for each 262 type. Different facilities can have other priority ranks, such 263 as a lumped load, and the load within each building can be 264 grouped based on their criticality.

266
A multi-stage controller that the MGMS dictate at the central 267 level and BEMS at the building level is formulated in this 268 paper to schedule/operate the various devices in the Banshee 269 microgrid system to ensure the resiliency of the dispersed 270 critical loads.

272
The optimization is performed to minimize the load shedding 273 by scheduling the various generation sources and switching 274 operations available in the microgrid system. When a general 275 blackout or partial power outage occurs in a microgrid, the 276 most critical concern is maintaining service to the priority 277 loads and the remaining customers affected. The objective 278 VOLUME 10, 2022 function is defined for the entire considered time horizon. The 279 optimization problem can be formulated by either (1) or (2), 280 subject to constraints in (5)-(39).

313
The AHP process prioritizes the load groups for curtail-314 ment by prioritizing load groups based on a priority vector.

315
The AHP consists of four steps: In pairwise comparison, the building load groups are 326 compared to each other based on relative importance.

327
An eigenvector created a scale of (1-9), which is assigned 328 for the pairwise building comparison. For example, if λ α 329 (building α) is more important than λ β (building β), it is 330 assigned a value of 9, and λ β must be less important than 331 λ α assigning a value of 1. From the eigenvectors a N L × N L 332 matrix A is formed. The judgment matrix, A is normalized by 333 multiplying the inverse of each column summation (W). The 334 normalized judgment matrix (A normal ) identifies the weights 335 of each criterion. After normalizing, the principal eigenvector 336 (A p ) can be calculated by averaging across the rows, which 337 determines the building priority as ratios. A p is a 1 × N L 338 matrix. The alternative decision matrix, B, is an a N L × N h 339 real matrix where numerical values (1-9) are assigned to the 340 relative importance of load groups in each building. As an 341 example, if the server group loads, x in building α (b xα ) 342 is more critical than the server group loads, x in building 343 The matrix B is normalized by 344 multiplying the inverse of each raw summation of B. The 345 decision priority matrix, ∅, can be calculated by multiplying 346 each element of the normalized alternative decision matrix 347 The problem is formulated as a MILP model that can be 357 effectively solved by commercial solvers. To restore a system 358 from a local outage state, the initial condition constraints 359 should be adequately configured to incorporate the infor-360 mation on the energization status and availability of each 361 component.

363
In this sub-section, the equality and inequality constraints 364 for optimizing the microgrid are discussed. The main control 365 variables which contribute to the operation performance are 366 tie-switches and DERs. The power-flow equations are inherently nonlinear due to 369 the loads modeled as constant P-Q, which requires several 370 iterations to find the solution. By modifying the load models, 371 the power-flow problem equations for a microgrid system can 372 be transformed into a linear system model. Simplified Dist-373 Flow equations have been widely used for radial distribution 374 systems based on the assumption that the nonlinear terms in 375 the DistFlow equations are much smaller than the linear terms 376 and can be ignored [32]. The following equations extend the 377 linear DistFlow equations using binary variables:  (11). Typically, V min is 0.95 p.u. and The active power is limited by the apparent power limit of the 419 inverter as given by (12) and (13).
The reactive power capability is determined by the present 423 active power and the power factor limit at each bus given 424 by (14) and (15)III-B4.
The P ESS i,t could either be a discharging or charging process 429 controlled by the decisions u ESS D i,t for discharging or u ESS C i,t for 430 charging, as represented by (16)(6) and (17)(6).
The state of charge (SoC) of each battery connected to a 435 bus at any time t, depends on its SoC at the previous time 436 step and its discharging/charging rate given by (18). The ESS can deliver exceptional flexibility for DSR because of 443 its capability of generating and absorbing power strategi-444 cally. A critical concern for ESS operation is to maintain the 445 residual energy, or SoC, during the charging and discharging 446 process. In this paper, it is assumed that both active and 447 reactive power can be dispatched for an ESS and that the SoC 448 is not affected by dispatching reactive power. For each ESS 449 i ∈ e at each bus, i step t ∈ T , the operational constraints 450 can be formulated by (20)- (22).
The amount of energy that DDGs can provide in an islanded 459 microgrid is limited by the available fuel reserves given 460 by (23).
The left-hand side of (21) is the total amount of energy 463 served by the DDG from the start of the emergency con-464 dition. On the right-hand side, E act DDG i,t is the upper limit 465 on the amount of energy that DDGs can provide during 466 emergency conditions. For convenience, these energy terms 467 are all measured by equivalent electric energy in kWh. 468 VOLUME 10, 2022 The conversion rate from the amount of fuel to electric energy 469 varies with machines and should be determined case by case. 470 As an example, the diesel generator generating 1-kWh elec-471 tric energy consumes about 0.086 gallons of diesel [20].

486
Equations (25)(21) To ensure that the energized system is operated in the tree 497 topology during the restoration process, the energized system 498 at each time step should be a connected graph and satisfy (29).
In this paper, there is one substation node, and several 501 black-start DERs are assumed to exist in the system. Con- levels. The building demand response is responsible for man-518 aging and maintaining energized as many groups of loads as 519 possible according to the power consumption limit set by the 520 MGMS. The load shedding is made according to the amount 521 of energy consumption that the building is required to reduce. 522 The h group of loads is dropped from the less to the most 523 critical one, and their drop is made until the requested power 524 consumption reduction is achieved. If the entire building load 525 is shed, then the constraint in (30) must be respected.
Considering the disintegration of loads into different 529 groups, the load groups can be shed by (31): where x L i,h is the controllable factor of the h th group load as a 536 percentage of the entire building, and the sum is less than 1, 537 signifying that the entire building cannot have controllable 538 loads. Once all buildings maintain their power consumption 539 less or equal to the limit set by the MGMS, the microgrid's 540 stable and safe operation is ensured.
where u CP i,y,t = {0,1}, represents the circuit breaker off 549 and on status of the EV charging station for one interval. 550 P CP i,y,t is the estimated charging power demand of the y th 551 EV in the next interval. It is obtained by using the current 552 SOC of the EV, the time period of the next interval, and 553 the predicted charging profile of the EV. Data mapping 554 with current SOC as the identification parameter is used 555 to determine the starting point in the predicted charging 556 profile. The number of data points equal to the number 557 of minutes in the next interval (30 min in this case) is 558 extracted to calculate the value of the power condition-559 ing system (PCS). The load model for predicting the 560 charging profile of the EV using initial SOC, final SOC, 561 and previous charging profile is described in [21]. The 562 entire building's EV power demand is then given by (33). 563 where N ev is the total number of EV charging stations 565 installed in the building.

566
• AC System Load Model: The power demand of one AC 567 system load is based on its motor power rating, circuit 568 breaker ON/OFF status, and motor operating speed con-569 cerning the VSD frequency as presented in (34).

610
A central BEMS coordinates with the MMS to schedule 611 the loads for real-time adjustment. The BEMS receives the 612 scheduled power, P L i,t determined by the optimization solu-613 tion described in Section III, which is provided to the patent-614 pending BEMS solution as thresholds, L i,t for each building. 615 The optimization is performed one time step ahead of the 616 scheduling time interval. In real-time, the forecasted schedule 617 of the buildings could be different from the actual power 618 consumption. The BEMS receives the thresholds from the 619 optimization solution and adjusts the group loads by switch-620 ing on/off in real-time to respect the threshold limits.  Next, the loads at each building level are measured in real-643 time, Pmeas L i,t and compared with the building threshold, L i,t . 644 If the building measurement is above the threshold, the loads 645 are curtailed to be under the threshold. This iterative cur-646 tailment process is repeated until measured building loads 647 become less than the threshold powers determined in the 648 scheduling period, with a gap of δ L i .

649
It may be the case that more power becomes available in 650 different time slots due to solar PV increase, which could 651 allow for turning on more loads. In such a scenario, the 652 next set of priority loads are scheduled in these time slots 653 until the threshold and maximum power gap are reduced. 654 When the building measurement is below the threshold, then 655 a load increment is performed to narrow the gap between the 656 threshold and the maximum power capacities of the next set 657 of priority loads until the gap is narrowed to δ L i . The reason for 658 considering the maximum loads is due to the unknown power 659 consumption of the groups until they are switched on, which 660 will also ensure scheduling without any chattering issues. For 661 each time step, t, the scheduled group loads are finalized as 662 xu L 1,1,t , .., xu L N L ,N h ,t . 663 VOLUME 10, 2022   The problem was formulated for a generic approach in 680 Eq. (1)-(38). Considering the Banshee network, the problem 681 is formulated with 98 nodes in N , and 22 lines in l . 682 The Banshee building loads are categorized in three different 683 priority ranges, critical, priority, and interruptible, as shown 684 in Table 1, [35]. The critical buildings are considered the 685 most important ones, which are mission-critical facilities and 686 strongly rely on power supply continuity. Priority buildings 687 are ranked after the critical ones and do not have a large 688 impact when de-energized. Finally, the interruptible buildings 689 can be de-energized for short periods without a considerable 690 impact.

691
Based on the original Banshee buildings' configuration, 692 the present paper proposes a breakdown structure that splits 693 the building's aggregated loads into five different desegre-694 gated groups of loads that compound h . Each building can 695 be disintegrated into load groups with different priorities. 696 For example, even though building P1 is less critical than 697 C1, P1 may have more important loads than some loads in 698 building C1. By having a granular structure, non-critical loads 699 in C1 can be disconnected and enable supply to P1 critical 700 loads. Each building can be restored in small load steps 701 according to the DER available capacity and the buildings 702 and group loads' rank position. Different weight factors are 703    The input to the MGMS is the distribution system parameter 724 and constraints and provides the hourly optimal solution for 725 the system based on real-time measurements, including DERs 726 dispatch and groups shedding commands. At the secondary 727 level, the central BEMS is deployed on a DELL central server 728 at the control station. The central BEMS is responsible for 729 monitoring the entire grid operation, collecting data, and for-730 warding hourly dispatch to the local controller at the primary 731 level. The grid operation includes control of breakers, active 732 and reactive power setpoints for DERs, voltage and frequency 733 regulation, and hourly consumption of the building. The pri-734 mary BEMS uses raspberry pi to implement the primary load 735 control algorithm at the building level. The output of the 736 algorithm is the dispatch commands to the building loads. 737 The load control algorithm is written in python programming 738 language and is deployed using the open-source agent-based 739 VOLTTRON platform. The VOLTTRON pub/sub message 740 framework handles data flow in the system. Each Raspberry 741 Pi hosts two VOLTTRON instances, one for each building 742 control. The local controller receives the building dispatch 743 command from the central BEMS and provides a granular 744 dispatch signal to the virtual building models in OPAL-RT. 745 The OPAL-RT OP5700 Real-time Digital Simulator com-746 pounds the base layer. OPAL-RT is well known for its high 747 capability to host large electrical systems. The entire Ban-748 shee Network, its DERs, and buildings were modeled using 749 RT-Lab software. The building model followed the proposed 750 granular structure presented in section IV, and each model 751 has a unique virtual IP address to exchange data with external 752 controllers.

754
The hierarchical control topology and its implementation 755 is presented in Figure 4. The tertiary level is responsible 756 VOLUME 10, 2022  Figure 6 illustrates the case studies. follow the priority order for emergency load shedding to 811 meet the power balance [34]. While considering the gen-812 eration schedule, the end time of the outage is not consid-813 ered, which forces the fuel generator and battery to produce 814 their maximum outputs at each time step. The generators 815 operate at their total capacity until their fuel is out, tripping 816 the device. Similarly, the battery runs until its discharges 817 completely. In this case, the buildings are again considered aggregated 820 lumped loads but under a one-week outage. While consider-821 ing the generation schedule, the end time of the outage is now 822 taken into consideration which will enable the fuel generator 823 to produce outputs conservatively for each time step. The fuel 824 generators and the battery will operate at optimal outputs. 825 This approach ensures that critical building loads are supplied 826 at the end of the predicated outage duration. This type of 827 outage duration estimation can be predicted based on past 828 events. In this case, the buildings are disintegrated with granular 831 group loads. The outage is unconstrained, and the generation 832 scheduling aims to supply as many loads as possible, similar 833 to Case 1.  Figure 9 shows a similar 865 analysis but for Case 4, which considers granular control 866 following groups' priority under different levels of criticality.

867
The graphs display the actual served and actual demand and 868 shed overall active power profiles for each group of loads.

869
As the priority rank increases from group 1 to 6, under shed 870 commands, the optimization algorithm aims to ensure that 871 group 6 followed by group 5 remain supplied by shedding 872 other non-critical load groups.

873
The Total Connected Load (TCL) factor is computed to 874 measure the levels of supplied loads during the period of 875 study. The TCL formulation for the overall building and 876 group load buildings is shown in (39) and (40), respectively. 877 The Total Actual Load (TAL) factor measures the total 880 actual load connected, u L i,t = 1, and the %shed calculates 881 the percentage of disconnected loads over the study period, 882 as shown in (41)   These results validate the proposed solution's capability 895 to ensure service continuity for critical loads by shedding a 896 large percentage of non-critical loads on each microgrid's 897 VOLUME 10, 2022 building. For the daily load analysis, from Table 3,   generation conservation algorithm is considered, interrupt-920 ible and priority buildings loads are shed to ensure the critical 921 buildings loads' service continuity during the entire 1-week 922 outage period, as shown in Case 3 from Fig. 10. When 923 the microgrid's buildings are categorized based on the same 924 groups of loads, the MGMS receives additional flexibility to 925 curtail non-critical load groups from all the buildings.

926
Similar to Case 1, in Case 2, all the generation is used 927 to meet the current load since the start of an outage. The 928 percentage of aggregated buildings shed for the outage period 929 is comparable to the percentage of group loads shed after the 930 13 th hour, as shown in Fig. 10. For Case 4, the interruptible 931 load groups are shed first to ensure the critical load groups' 932 service continuity during the outage period. In this scenario, 933 groups 4 and 5 remain in service, as shown in Fig. 10. While in an aggregate scenario, buildings must be completely 936 de-energized to respect the power balance constraints, a gran-937 ular capability of control enables a higher level of flexibil-938 ity and precision for demand response shedding. Smaller 939 amounts of loads can be disconnected to respect the avail-940 able levels of energy. Buildings' group loads are turned off 941   fuel and storage energy is used at the beginning to supply 949 critical and non-critical loads, the system's reliability may 950 be compromised as the outage perpetuates. However, the 951 outage duration can be predicted based on previous events. 952 This information can be essential to scheduling the load and 953 generation to supply critical loads over the outage period.

955
This paper has proposed a resilience-oriented optimization 956 strategy for networked microgrids focusing on building-level 957 management during emergency conditions. The main contri-958 butions of this research are to integrate the microgrid opera-959 tions at the tertiary level with the BEMS, the central BEMS 960 at the secondary level, and the individual building controllers 961 at the primary level. This type of integrated solution is cur-962 rently not available in the market but is highly required for 963 critical facilities operation with reliable conditions during 964 outage periods. All proposed control levels work interac-965 tively in real-time to make decisions based on measurements 966 and grid conditions. High-level controls take decisions on 967 the grid benefit, and low-level controls optimize loads and 968 DERs, respecting higher-hierarchical commands. The pro-969 posed solution has been implemented on real-time prototype 970 platforms and validated with a real-time simulator. With 971 increasing building flexibilities which include HVAC, light-972 ing, plug-load controllers, EV chargers, water heaters, etc., its 973 integration to BEMS platforms will give microgrid operations 974 the much-needed flexibility to curtail several non-critical 975 loads distributed in different buildings.   It can be concluded that the proposed resilience-oriented 991 optimization strategy can reduce load shedding considerably 992 at the cost of a minute increase in the operation cost. service restoration for advanced distribution systems and microgrids,'' degree in mechanical and manufacturing engineer- 1179 ing from the University of Rhuna, Srilanka, and the 1180 doctorate degree in mechanical engineering with 1181 a focus on dynamical systems and control from 1182 Texas Tech University.

1183
He is the Lead Research Control Engineer at 1184 Group NIRE. Throughout his studies and indus-1185 trial internship, he has obtained over seven years 1186 of experience studying industrial network control 1187 systems, SCADA, and mobile robots. Recently, he led the Design Engineer-1188 ing Team to develop the SCADA system using Laboratory-view, Qt, and 1189 node-Red programming platforms for TTU, GLEAMM, 750kw, micro-grid, 1190 Lubbock, TX, USA, to facilitate the next generation micro-grid research. 1191 He also piloted the Design Engineering Team to develop the data monitor-1192 ing for TTU HUMS sustainable smart home project, Junction, TX, USA, 1193 to enable high precision data collection and monitoring for sustainable 1194 living research at TTU. He also worked on developing SCADA systems 1195 for laboratory-scale microgrid testbeds at Texas Tech University. His current 1196 research interest includes developing energy management system (EMS) for 1197 critical infrastructure.