Multi-agent system based active distribution networks

This thesis gives a particular vision of the future power delivery system with its main requirements. An investigation of suitable concepts and technologies which creates a road map forward the smart grid has been carried out. They should meet the requirements on sustainability, efficiency, flexibility and intelligence. The so called Active Distribution Network (ADN) is introduced as an important element of the future power delivery system. With an open architecture, the ADN is designed to integrate various types of networks, i.e., MicroGrid or Autonomous Network, and different forms of operation, i.e., islanding or interconnection. By enabling an additional local control layer, these so called cells are able to reconfigure, manage local faults, support voltage regulation, or manage power flow. Furthermore, the Multi-Agent System (MAS) concept is regarded as a potential technology to cope with the anticipated challenges of future grid operation. Analysis of benefits and challenges of implementing MAS shows that it is a suitable technology for a complex and highly dynamic operation and open architecture as the ADN. By taking advantages of the MAS technology, the AND is expected to fully enable distributed monitoring and control functions. This MAS-based ADN focuses mainly on control strategies and communication topologies for the distribution systems. The transition to the proposed concept does not require an intensive physical change to the existing infrastructure. The main point is that inside the MAS-based ADN, loads and generators interact with each other and the outside world. This infrastructure can be built up of several cells (local areas) that are able to operate autonomously by an additional agent-based control layer. The ADN adapts a MAS hierarchical control structure in which each agent handles three functional layers of management, coordination, and execution. In the operational structure, the ADN addresses two main function parts: Distributed State Estimation (DSE) to analyze the network topology, compute the state estimation, and detect bad data; and Local Control Scheduling (LCS) to establish the control set points for voltage coordination and power flow management. Under the distributed context of the controls, an appropriate method for DSE is proposed. The method takes advantage of the MAS technology to compute iteratively the local state variables through neighbor data measurements. Although using the classical Weighted Least Square (WLS) as a core, the proposed algorithm based on an agent environment distributes drastically computation burden to subtasks of state estimation with only two interactive buses and an interconnection line in between. The accuracy and complexity of the proposed estimation are investigated through both off-line and on-line simulations. Distributed and parallel working of processors improves significantly the computation time. This estimation is also suitable for a meshed configuration of the ADN, which includes more than one interconnection between each pair of the cells. Depending on the availability of a communication infrastructure, it is able to work locally inside the cells or globally for the whole ADN. As a part of the LCS, the voltage control function is investigated in both steady-state and dynamic environments. The autonomous voltage control within each network area (cell) can be deployed by a combination of active and reactive power support of distributed generation (DG). The coordinated voltage control defines the optimal tap setting of the on-load tap changer (OLTC) while comparing amounts of control actions in each area. Based on the sensitivity factors, these negotiations are thoroughly supported in the distributed environment of the MAS platform. To verify the proposed method, both steady-state and dynamic simulations are developed. Simulation results show that the proposed function helps to integrate more DG while mitigating voltage violation effectively. The optimal solution can be reached within a small number of calculation iterations. It opens a possibility to apply the proposed method as an on-line application. Furthermore, a distributed approach for the power flow management function is developed. By converting the power network to a represented graph, the optimal power flow is understood as the well-known minimum cost flow problem. Two fundamental solutions for the minimum cost flow, i.e., the Successive Shortest Path (SSP) algorithm and the Cost-Scaling Push-Relabel (CS-PR) algorithm, are introduced. The SSP algorithm is augmenting the power flow along the shortest path until reaching the capacity of at least one edge. After updating the flow, it finds another shortest path and augments the flow again. The CS-PR algorithm approaches the problem in a different way which is scaling cost and pushing as much flow as possible at each active node. Simulations of both meshed and radial test networks are developed to compare their performances in various network conditions. Simulation results show that the two methods can allow both generation and power flow controller devices to operate optimally. In the radial test network, the CS-PR needs less computation effort represented by a number of exchanged messages among the MAS platform than the SSP. Their performances in the meshed network are, however, almost the same. Last but not least, this novel concept of MAS-based AND is verified under a laboratory environment. The lab set-up separates some local network areas by using a three-inverter system. The MAS platform is created on different computers and is able to retrieve data from and to hardware components, i.e., the three-inverter system. In this set-up, a configuration of the power router is established in a combination of the three-inverter system with the MAS platform. Three control functions of the inverters, AC voltage control, DC bus voltage control, and PQ control, are developed in a Simulink diagram. By assigning suitable operation modes for the inverters, the set-up successfully experiments on synchronizing and disconnecting a cell to the rest of the grid. In the MAS platform, an obvious power routing strategy is executed to optimally manage power flow in the lab set-up. The results show that the proposed concept of the ADN with the power router interface works well and can be used to manage electrical networks with distributed generation and controllable loads, leading to active networks