The core functionality, which has been demonstrated with continuous operation of about 150 devices during 1 year since March 2015, consists of sampling every second the measurement data of each GridBox device and collecting this information in quasi real-time on a GridBox Master. Based on this functionality, two types of applications are possible. On the one hand, real-time applications, so called smart grid applications (SGAs), are executed with the set of information available every second. These include power system state estimation, topology estimation, monitoring and prosumer control. On the other hand, ex post applications that require archived measurement data can be run. The interplay of the SGAs and the archive service within the GridBox system is illustrated in Fig. 3. Here, the measurement accumulator and the governor services provide an interface for the SGAs to receive measurements from and send control signals to the GridBox devices, respectively. Topology estimation of Sect. 4.1 tries to determine the connectivity graph or even the entire model of the corresponding grid. Assuming that a grid model is given, the state estimation of Sect. 4.2 completes and adjusts the GridBox measurements which can be inaccurate or missing altogether. The estimates of the state estimation can then be used to formulate an optimization problem. The solution to this optimization problem determines a set of control signals which, after sending them to the corresponding GridBox devices, improve the overall grid state with respect to a certain objective. This optimal power flow control SGA is discussed in Sect. 4.3. Furthermore, a monitoring SGA and the data archive service are presented in Sects. 4.4 and 4.5, respectively.
Automated topology identification is a subsequent application which will be tested during the pilot project. The objective is to draw a connectivity graph, i.e. to determine which nodes are connected with each other. Therefore, an algorithm matches node currents on both sides of a line. Especially in a meshed distribution network, the connectivity graph may be subject to frequent changes by switching operations. Even though a distribution system operator normally possesses at least one network diagram for planning and operation purposes, the automated identification can replace manual interventions for the initial input of network data or updates in the case of topology switches. Adding node voltages to the current phasor data used in the topology identification, the electrical parameters of the lines can be obtained so that load flow calculations can be performed. For a discussion of algorithm approaches and results, the reader is referred to [6, 7].
The elementary real-time application consists of a linear three-phase state estimator based on concepts presented in [8, 9] capable of detecting and treating bad data such as measurement deviations and of handling missing data. Missing data can be of temporary nature in case of communication interruption. However, more relevant are nodes that are not measured at all. The line model used is shown in Fig. 4.
Besides technical reasons such as inaccessible cable junctions in the ground, it is also an economical motivation to deploy only as many GridBoxes as necessary. In both pilot networks, most accessible nodes are equipped with a GridBox device in order to be able to evaluate the confidence metric of the state estimator by deliberately considering only a subset of the available measurement points. Section 6.1 explains the outcomes of this analysis. The state estimation associates to each node estimated values and confidence intervals for current and voltage.
Optimal power flow control
In order to dispatch all controllable prosumers (actors) in an optimal way, an application was developed that solves the three-phase optimal power flow (OPF) problem for the entire grid region with objectives like voltage stability or reducing line or transformer strain at multiple locations in the grid. Further goals could be minimization of grid losses. With that, the DSO will be able to choose among several objective functions.
In order to be reactive enough for voltage stability improvement, the computational load needs to be kept low. This has been achieved by implementing a single-timestep OPF using sequential linear programming (after conducting some preliminary studies based on semidefinite programming, see ). Actors with storage capabilities are often controlled using model predictive control (MPC) . As this project focuses on the geographical extent as well as high control reactivity in the order of seconds, such a computation-heavy approach has been neglected.
In the pilot grids, the optimizer can control PV plants by curtailing active power or requesting a certain power factor. A BESS located in the ewz pilot grid can be controlled in active as well as reactive power. Water boilers as well as heatpumps can be switched on or off in order to shift their load.
The actual state of the distribution network is calculated and evaluated against the permissible operational limits at every instant. To visualize the actual state of the grid as well as constraint violations and historic data, a cockpit has been developed as a web-service as shown in Fig. 5. Three-phase voltages as well as power flows are animated in real time, providing the DSO with a today lacking level of transparency for the lower grid regions.
The second major set of possible functionalities can be seen as a by-product of the aforementioned real-time applications. An archive with measured values, like synchrophasors and harmonics of voltages and currents as well as outputs of the real-time algorithms, like estimated values, switching states and optimizer set-point values, can be used to improve existing algorithms and develop new applications with respect to realistic conditions. The synchrophasor measurements could be used to develop sophisticated state estimation algorithms based on machine-learning techniques. As another example, the load and voltage profiles could lead to more efficient network planning. Furthermore, the capability of the GridBox to measure synchronized phasors would allow the implementation of innovative future applications such as fault detection (i.e. locate the exact place where a fault has appeared), transient analysis, detection of unintentional islanding or management and operation of micro-grids (i.e. reconnect a part of the grid without injecting disturbances).
The GridBox platform is designed as open as possible in order to be able to address future questions that are not yet foreseeable today or that require longterm data.