Scenario-Generalized Multi-Stage Dynamic Programming for Online Dispatch of Distribution Networks via Universal Value Function Learning

The advent of distributed generation (DG) units has posed many problems to the planning and operation of distribution networks (DS). The selection of the “best” site for the DG installation is a major problem in this context. Different strategies and methods have been developed in the recent years which deal with this problem. However, Complexity of solutions and inadequate manipulation have hindered their utilization by DS planners. This paper presents a robust and practical tool which helps the distribution network planner in getting the best possible location of DG unit taking into account both reliability of supply and voltage sag problems. The method suggested uses knowledge-base expert system (ES) to help in selecting the appropriate locations. The method suggested is based on economic and technical considerations. It investigates with economic loss due to both voltage sag and forced outage of DS components. It concerns with the effect of location of the added DG on the overall network performance during both steady-state when a forced outage occurring on one or more component of the DS and transients after sudden short-circuit. Using ES avoids mathematical complications of solving complex analytical model. By judicious rule-base ES, selection of optimum locations becomes clear and simple. The suggested method is explained and implemented through practical distribution network and its effectiveness is tested and discussed.

With large-scale distributed energy resources (DERs) being integrated into distribution networks (DNs), the coordinated optimal power flow (OPF) of high voltage (HV) and medium voltage (MV) DNs becomes challenging due to enhanced uncertainties and nonlinearity. This paper presents a hierarchical distributed chance-constrained OPF model for HV and MV DNs based on a nonlinear adaptive data-driven method. The HV master problem aims to minimize the overall operational costs, and is transformed into mixed-integer convex problem via second order cone programming. A model-free chance-constrained optimization sub-problem is established to tackle inaccurate model parameters issues and DER uncertainties in MV DNs and realize rapid response to control schemes from HV DN. The nonlinearity of power flow is approximately depicted by a data-driven incomplete dimension lifting power flow (IDL-PF) constraint, which is integrated into the MV DNs optimization model. A novel multi-parameter programming (MPP)-based distributed algorithm is proposed to solve the coordinated OPF model, which can significantly accelerate the iterative speed between the master and sub-problems. Finally, the convergence, efficiency and optimality of the proposed method are validated by numerous simulations in a practical DNs in China.

Dual-layer peer-to-peer energy trading method with multiple SOPs pricing in distribution networks

In recent years, with the steady growth of load demand in distribution networks, the fluctuation and uncertainty of power loads have significantly increased. Meanwhile, the rising penetration of photovoltaic generation has further exacerbated the challenges of power system accommodation capability. To enhance photovoltaic accommodation capability and realize the secure and economic operation of distribution networks, a multi-time scale hierarchical coordinated optimization operation strategy for distribution networks with aggregated distributed energy storage has been proposed. First, the regulation requirements of aggregated distributed energy storage are analyzed, and a distributed energy storage aggregation model is established based on an inner approximate Minkowski Sum. Subsequently, a multi-time scale optimization operation model considering source-load uncertainties for day-ahead, intra-day, and real-time stages is developed based on a stepped carbon emission cost model. Then, a power allocation method within the aggregated distributed energy storage based on the water-filling algorithm is presented. Finally, a practical distribution network in a demonstration county in China is used as a case study to validate the proposed method. The results demonstrate that the proposed strategy effectively reduces system operation costs while improving photovoltaic accommodation capacity and enhancing the reliability of system operation.

Utilization of active power filters (APFs) is the most efficient method to reduce harmonic pollution in distribution networks. Previous approaches utilized APFs in integrated control schemes based on broad data-gathering systems. Since a broad data-gathering system is not available in most practical distribution networks, previously proposed approaches may not readily be implemented. This paper presents the utilization of stand-alone controlled APFs (SACAPFs) in radial distribution networks. Utilizing APFs with a stand-alone control system decreases implementation costs and complexity by making them autonomous and independent of integrated control systems, which are complicated and expensive in practical applications. In this paper, a single SACAPF is modeled as a dependent current source where its injection current is equal in amplitude but opposite in phase compared to the harmonic content of the current passing through the point of common coupling (reference current). Due to the presence of both linear and nonlinear loads in the distribution network, the reference current changes after injection by SACAPF, so it is necessary to modify the injection current until reaching a constant value in the reference current. This is considered via an iterative procedure in the modeling scheme. Operation of multiple SACAPFs is handled using a backward procedure based on a priority list. Simulation results on an IEEE 18-bus test system show the proper operation of the stand-alone control systems for both single and multiple SACAPF implementation. Furthermore, optimal allocation of the proposed SACAPFs is performed in an IEEE 33-bus test network and a 9-bus test network, and the results are discussed and compared with the allocation of integrated control system APFs.

Peer-to-Peer electricity trading of interconnected flexible distribution networks based on Non-Cooperative games

Regarding the increase in electrical power demands, distributed generations (DGs) from renewable energy resources are becoming essential for electrical generation systems. However, there are no specific requirements for DGs allocation in a distribution network. The improper installation of the DGs can cause crucial issues to electric utilities and customers, such as the increase in energy losses and the improper voltage drop in the networks. In this paper, the optimal distribution network reconfiguration implemented with a tie line and a capacitor is proposed to enhance system efficiency and mitigate the impacts of the improper DGs installation. The optimal allocation of the tie line and the capacitor is determined using an improved particle swarm optimization (I-PSO) technique to minimize the energy losses and improve the voltage profile. The proposed network reconfiguration approach is implemented into a practical distribution network with a high installed capacity of the biomass DG unit from Kamphaeng Phet province, Thailand. The simulation test results show that the proposed approach can significantly reduce not only the energy losses but also the voltage drop in the network.

We study the problem of directly optimizing arbitrary non-differentiable task evaluation metrics such as misclassification rate and recall. Our method, named MetricOpt, operates in a black-box setting where the computational details of the target metric are unknown. We achieve this by learning a differentiable value function, which maps compact task-specific model parameters to metric observations. The learned value function is easily pluggable into existing optimizers like SGD and Adam, and is effective for rapidly finetuning a pre-trained model. This leads to consistent improvements since the value function provides effective metric supervision during finetuning, and helps to correct the potential bias of loss-only supervision. MetricOpt achieves state-of-the-art performance on a variety of metrics for (image) classification, image retrieval and object detection. Solid benefits are found over competing methods, which often involve complex loss design or adaptation. MetricOpt also generalizes well to new tasks and model architectures.

Optimal dispatch of unbalanced distribution networks with phase-changing soft open points based on safe reinforcement learning

The network reconfiguration technique has been used to reduce network losses, balance transformer loading, and restore the power supply of the outage area in distribution networks. With a large number of uncontrollable distributed generators (DGs) added to distribution networks, the reconfiguration technique can be applied to enhance the available delivery capability (ADC) of distribution networks. This paper presents a two-stage methodology for determining optimal network topologies for improving the ADC of distribution networks for supporting more renewable energies. The first stage is a group-based binary particles swarm optimization (BPSO), which takes advantage of the BPSO's global capability while improving its slowness in finding optimal solutions. The second stage is a hybrid of greedy-based and domain-knowledge-based methods. When applied to multiple scenarios of DGs, the proposed methodology can effectively deal with the uncertainty of the DGs' outputs and time-varying loading conditions. As a byproduct, the optimal network topology found by the proposed methodology also decreases the network power losses due to the increase of the ADC of the distribution networks. The IEEE 123-bus test network and a practical 1001-node distribution network are used to verify the proposed methodology, and the numerical studies demonstrate the effectiveness of the proposed methodology in greatly increasing the ADC via optimal network reconfigurations.

The modern active distribution systems necessitate integrating storage systems, thereby facilitating the large-scale proliferation of photovoltaic (PV) energy resources. This further calls for the optimal planning of energy storage systems, satisfying all the operational and economic constraints. This article describes an exhaustive storage integration method, deeming the life cycle of the battery energy storage, the uncertainty of load and PV output, and the islanded mode of operation of the system. A two-stage mixed-integer linear programming problem is formulated that determines the capacity and the number of discharge cycles of the batteries in the first stage. The lifetime of the battery based on the partial depth of discharge is analyzed in the second stage. Furthermore, the uncertainty and variability of PV and demand are taken into account through probabilistic analysis and time-period clustering. The method is validated on a standard 33-bus radial distribution network for the allocation of distributed lithium-ion batteries. Also, the method’s scalability is validated on a practical Indian distribution network and a 141-bus distribution network of metropolitan area of Caracas with distributed PV installations on various nodes.

In this paper a new and efficient method for solving the load flow problem of a distribution system is discussed. It is based on network topology, basic circuit laws and power summation technique. The main contribution of this paper is: (i) It proposes a new, efficient load flow method for radial and weakly meshed distribution systems, (ii) It evaluates the impact of load models, different values of X/R ratios, load growth for future expansion and tolerance levels, (iii) It analyses the impact of number of loops on weakly meshed distribution systems, (iv) and also gives comparison of radial and weakly meshed distribution system. The results are obtained for voltage values, total real and reactive power losses, time of computation, and number of iterations required. Computer program coded to implement this power flow solution scheme in MATLAB and successfully applied to several practical distribution networks with radial and weakly meshed structure. Effectiveness of the proposed load flow method has been presented on IEEE 33bus radial and meshed distribution networks.

A two-stage stochastic programming approach is pursued to optimally place and size photovoltaic (PV) inverters in a radial distribution network under solar irradiance and load uncertainties. First-stage variables include binary PV unit placement as well as continuous real and apparent power capacities of the inverters. Second-stage decisions comprise reactive power compensation, power flows, and nodal voltages, which are determined adaptive to the uncertainty. The objective is to minimize installation cost and expected thermal losses on the network. The optimization problem is formulated as a mixed-integer second-order cone program and is tested on a practical distribution network.

Many modern robotic systems such as multi-robot systems and manipulators exhibit redundancy, a property owing to which they are capable of executing multiple tasks. This work proposes a novel method, based on the Reinforcement Learning (RL) paradigm, to train redundant robots to be able to execute multiple tasks concurrently. Our approach differs from typical multi-objective RL methods insofar as the learned tasks can be combined and executed in possibly time-varying prioritized stacks. We do so by first defining a notion of task independence between learned value functions. We then use our definition of task independence to propose a cost functional that encourages a policy, based on an approximated value function, to accomplish its control objective while minimally interfering with the execution of higher priority tasks. This allows us to train a set of control policies that can be executed simultaneously. We also introduce a version of fitted value iteration to learn to approximate our proposed cost functional efficiently. We demonstrate our approach on several scenarios and robotic systems.

Unbalance active distribution network (UADN) is the combination of unbalance distribution network (UDN) and distributed generations (DGs) such as gas generator, wind turbines, fuel cell, photovoltaic system, etc. Regarding to unbalance structure, different kinds of DGs models and bidirectional power flow, some network parameters such as amplitude voltage, voltage unbalance Factor (VUF), and power losses of network will alter accordingly. A suitable power flow analysis will provide a platform to analyse these behaviours in steady state condition. This paper offers appropriate method for handling DGs in a Power flow program which is developed using MATLAB® for practical unbalanced distribution network at Iran and investigate the impact of DG location and DG control mode (as PV and PQ mode) on voltage profile, reducing power loss and VUF. (4 pages)