Distribution network planning considering the integration of demand-side resources

Maximizing the potential value of demand-side resources in terms of network reliability and economy is a considerable challenge because of their time-varying and potentially complex operational characteristics. It is urgent to take the interactive relationship between supply and demand side into consideration to improve the efficiency and practicability of the distribution network planning and enhance the utilization of existing equipment. In this paper, the uncertainty of DG output and the load is first modelled, and the seasonal label information is added to the model to generate the planning-oriented scenarios of DG output and the load. Considering the integration of demand-side resources, a bi-level optimal planning model is established to minimize the annual total cost of the distribution network. The proposed model is tested on the IEEE 33-node distribution system to illustrate its feasibility and validity in distribution network planning with consideration of the integration of demand-side resources.


Introduction
The chief contribution of demand-side resources in distribution network planning is the expectation of a significant improvement in the efficiency of network operation through the associated increase in the volume of reserve capacity available to compete with supply-side capacity 1 . However, the increasing penetration of demandside resources, such as demand response (DR), electric vehicle (EV) and distributed generation (DG), has brought substantial uncertainties to the distribution network. It has been recognized that sole changes on the supply side, such as substation expansions and network reconstructions, are insufficient in the long run and fail to reap full environmental and economic benefits 2 . Therefore, it is necessary to address the issues raised by the integration of demand-side resources to achieve their full accommodation and efficient utilization.
Currently, research on distribution network planning mainly focuses on individual units of demand-side resources. In reference 3, an integrated methodology that considers DG and DR as options for planning distribution networks is researched. However, this model downplays the coordinate interaction between demand-side resources allocation and distribution network planning. As a result, the model merely contributes to reflect the impact of demand-side resources on load profiles. The same is true for reference 4, in which only a general form of demandside resources has been considered. Reference 5 analyses the role of DG penetration, storage and incentive-based DR on distribution network planning, but the impact of EV goes unacknowledged. Therefore, in practical distribution network planning, it is of great importance to comprehensively consider the influence of all or major components of demand-side resources in order to build a collaborative planning model 6 .
The remainder of this paper is organized as follows. In Section 2, typical daily curves of photovoltaic output and load under multi-scenarios are proposed to deal with the problem of DG output uncertainty and seasonal variation of load. In Section 3, a bi-level coordinated distribution network planning model considering the integration of demand-side resources is built. Section 4 presents a case study to evaluate the proposed model. Conclusion is drawn in Section 5 in the interest of distribution network planning.

Multi-scenarios generation
In view of the fluctuation characteristics of DG output and the seasonal variation of load, the typical daily curves of photovoltaic (PV) output and load under multi-scenarios are generated to better capture their dynamic patterns in both temporal and spatial dimensions as well as accurately learning the probability distributions. Furthermore, the curves serve as a useful scenario set for the bi-level planning model built in the next section.
In this paper, the generative adversarial network (GAN) model, which is based on two interconnected deep neural networks, is adopted for scenario generation. New scenarios are directly generated based on historical data, without explicitly specifying a model or fitting probability distributions. In this way, the downside of existing methods based on probabilistic models is avoided.
As shown in figure 1, 12 PV output scenarios of four

Bi-level planning model considering demand-side resources
The distribution network planning with demand-side resources is a nested optimization problem characterised by the coupling of planning and operation. The common characteristics of network planning problem are stated as follows: (1) Each level independently optimises its own objective function. Decisions made at the planning level are the allocation schemes of demand-side resources with the objective of cost optimization, and 'decisions' made at the operational level are the operational parameters with the objective of reliability optimization.
(2) Objective function of each level is partially determined by variables at the other level. The operational level determines network reliability after allocation schemes are devised at the planning level.
To sum up, the two levels are interactive and exist within a predominantly hierarchical structure. Therefore, in order to investigate the full potential of demand-side resources, a bi-level distribution network planning model considering the integration of demand-side resources is presented.

Objective function
3.1.1 Upper planning model. The objective function of the upper planning model, as is shown in equation (1), is to minimize the total cost over the planning horizon, including the investment and operational cost of the distribution network, and the environmental benefits after the integration of demand-side resources.
where ec C is the annual total cost of the distribution network. r is the average social discount rate, which is generally taken as 8% in China. y is the planning horizon.  (2) Operational cost: where n is the number of scenarios. i f is the daily operational cost of scenario i, the formula of which is shown in equation (4). i  is the occurrence probability where buy C is the electricity price of the upper grid. , j t e is the load of node j at time t.

Constraints
The constraints contain not only traditional electrical restrictions (for instance, limitation of DG penetration, power flow constraint, etc.), but also restrictions including demand side management and network reconfiguration.
(1) Power flow constraint where k P and k Q are the active and reactive power at node k.

Algorithms for the solution of the bi-level planning model
The proposed integrated planning model is a nonlinear optimization problem of discrete variables, which can be effectively solved by the particle swarm optimization algorithm (PSO) 7 . The obtained demand-side resources allocation scheme from the upper level serves as a good initial point for reliability analysis in the lower level. Likewise, the lower level returns the daily operational cost of the distribution network to the upper for total cost calculation. Flowchart of the proposed bi-level planning model considering demand-side resources is shown in figure 4.

Case study
The effectiveness and advantages of the model and algorithm mentioned above, and the necessity of considering the demand-side resources are verified by simulation cases based on the IEEE 33-node distribution system. Figure 5. Structure of the IEEE 33-node distribution system.

Parameters
As stated by The code of planning and design of urban electric network 8 , the parameters of 35kV / 10kV substations are shown in table 1, in which the capacity of the substations before expansion is 31.5MVA. Feeder and DG parameters are shown in table 2 and 3. Table 4 presents the electricity pricing strategy according to the data provided by a power company.  Table 4. Electricity pricing strategy.
(2) To facilitate the study on the spatial characteristics of EV charging load, the IEEE 33-node distribution system is divided into three parts, namely the residential, industrial and commercial districts. The EV charging stations are installed at node 12, 19 and 29. Figure 6 shows the daily charging load profiles of three types of EV.
(4) The conventional load grows at an annual rate of 5%.

Results
Five different planning schemes are devised for comparison: (1) Scheme 1 is the optimum scheme which adopts the proposed bi-level planning model considering the integration of demand-side resources.
(2) On the basis of Scheme 1, Scheme 2-4 respectively excludes the access of DR, EV and DG.
(3) Scheme 5 marks a conventional distribution network with no demand-side resources integrated.
The results are shown in table 5. The comparison of cost and reliability among schemes is shown in table 6.

Analysis
(1) Compared with Scheme 2, the annual total cost of the distribution network under Scheme 1 drops by 7.73%, which indicates that DR can bring more economic benefits to the distribution network with lower implementation cost, and alleviate the problem of power shortage during peak hours. Furthermore, the implementation of IL in nodes with heavier load achieves higher economic benefits.
(2) Compared with Scheme 3, the annual environmental benefits under Scheme 1 increases by 49.45%, which suggests that EV enjoys a greater level of environmental friendliness. In the meantime, from the perspective of income, Scheme 1 proves to be more profitable. However, the peak load escalates after the access of EV, which calls for higher substation and feeder capacity.
(3) Granted that there is a certain amount of construction, operation and maintenance cost involved in the access of DG, but the annual total cost of the distribution network under Scheme 1 drops by 13.33% compared with Scheme 4. This is because the access of DG effectively lightens the need for a network upgrade. In this way, the cost of substation and feeder expansion is saved while the environmental benefits increase.

Conclusion
This paper addresses the issue of distribution network planning considering demand-side resources. First, multiple scenarios are generated to depict the uncertainty of DG output and the load. Based on this, a bi-level planning model is built with consideration of the potential benefits of demand-side resources at operation and economic levels as well as reliability enhancement. At last, the feasibility of the proposed model is confirmed through a case study and the following conclusions are drawn: (1) The comparison between the proposed planning scheme with its counterparts shows that demand-side resources can provide additional energy and capacity resources without installing new generation infrastructure, namely a network upgrade. Also, the integration of demand-side resources can enhance distribution network reliability and security.
(2) Demand-side resources produce higher economic and environmental benefits at a lower cost. Implementation of distribution network planning with no account of demand-side resources integration is likely to produce a network reconstruction scheme that may disagree with the real situation.