MARCO TEÓRICO Y CONTEXTUAL DE REFERENCIA
2 EL ORIGEN DEL NUEVO PARADIGMA Y LOS RETOS PARA ENFRENTARLO
2.3 La gestión del conocimiento
In order to analyse the impact rooftop PV system on the existing distribution network, methodology shown in Figure 4.2 is adopted. The methodology consists of two types of analysis, i.e. time-series and probabilistic analysis.
Figure 4.1: Basic modelling approach
4.2.1 Time-series analysis:
Prior to the PV integration analysis, a base case is determined to establish the initial state of the test network. The base case is conducted on an annual time-series load flow simulation and it is assumed that there are no PV systems connected. Once the initial state of the network is established, the annual time-series load flow simulation with installable PV capacity is conducted to determine critical point (greatest load PV generation difference) for further analysis. When the critical point is identified, probabilistic impact assessment described in subsection 4.2.2 is conducted for that particular point.
Time Series Simulation procedure
1. Acquire annual load demand and PV generation and save as .csv file with time stamp
2. Create load and PV characteristic in DIgSILENT PowerFactory as described in AppendixA.3
3. Run a load flow using a Quasi-Dynamic command in PoweFactory
4. Plot results (Voltage, voltage unbalance factor and equipment loading) for te simulation duration
4.2.2 Probabilistic analysis
There is high level of uncertainty in customer owned PV uptake and thus require a randomised analysis. In this study, Monte Carlo Simulations (MCS) are conducted to generate random variables to effect location and size of PV system as depicted in . It is assumed that all households are PV candidates with equal chances of installing PV
Base Case (no-PV)
Load flow computation per PV allocation
Figure 4.2: Proposed impact assessment methodology
system. To generate random locations and number of PV systems in a scenario, PV out of service parameter in Powerfactory is assigned a switching characteristic based on random variables following uniform probability distribution function (PDF) of 0
and 1 (where, 0 = in service and 1 = out of service). PV size is varied based on a potential rooftop PV limited by available roof space. PV scaling factor parameter is assigned a characteristic based on random variables following a beta PDF between 0 and 1.
These input variables are generated in Microsoft excel and saved as .csv file which
Random
Monte Carlo Simulation System modeling and Simulation Results Analysis
.csv .xlsx
Load demand Voltage
unbalance
Figure 4.3: Probabilistic Load flow
can be read by Powerfactory. Each scenario is assigned a time step with a total step equal to number of MCS iterations and 2500 iterations was adopted for this study.
Once .csv file is updated, load flow simulations are conducted for all possible scenarios in Powerfactory and results (voltage and equipment loading) are stored for further analysis. Load flows can either be balanced or unbalanced depending on how the test system was modelled. The output peak values per scenario are plotted into scatter plot for further analysis.
Probabilistic Simulation procedure
1. Identify instance of excess generation (Pnet = Pgeneration− Pload) 2. save as .csv file with time stamp
3. Create load and PV characteristic as described in Appendix A.3 4. Run a load flow using a Quasi-Dynamic command in PoweFactory
4.2.3 Variables compilation
As presented in Figure4.3, PV location, size and load demand are the three variables to be randomly varied based on PDF. Random number generator is critical in cre-ating random variables and Microsoft Excel is used in this study. A standard excel function RAND() is used to generate a random number between 0 and 1. These ran-dom numbers are generated based on software internal algorithm, which makes them pseudo-random. However, for the balance of this work, the term ’Random-number’ is used to refer to this pseudo-random number.
Load
Customer owned PV capacity is considered to be random as its adoption is reliant many factors that can not be modelled with certainty. In this study, PV capacity is varied by applying a scaling factor to the rated capacity of individual customer load.
f (x) = BET A.IN V (RAN D(), α, β, [A], [B]) (4.1) where
• α and β are beta distribution parameters
• [A] is the lower bound
• [B] is the upper bound
Probability distribution function for a specific load will be derived from historic data of a given network. Equation4.1set out to generate stochastic load inputs that follow a PDF that represent load population.
PV Allocation
The adoption of customer owned PV is governed by many factors as detailed in lit-erature and their adoption is random. In order to represent this randomness, PV systems are switched in and out of service. to effect a uniform distribution with equal chances of either 0 or 1, the formula (4.2) was applied in excel to generate random PV allocations.
f (x) = IF ∗ (RAN D() < 0.5, 0, 1) (4.2) as a result, all random number below 0.5 will be rounded off to 0 and the applicable
the applicable PV will be switched out. This allocation is independent of the state of other PVs in the feeder.
PV capacity
Similar approach used for load modelling is also adopted in modelling input parameters for customer owned PV capacity. is considered to be random as its adoption is reliant many factors that can not be modelled with certainty. In this study, PV capacity is varied by applying a scaling factor to the rated capacity of individual PV system.
f (x) = BET A.IN V (RAN D(), α, β, [A], [B]) (4.3) where
• α and β are beta distribution parameters
• [A] is the lower bound
• [B] is the upper bound
Potential rooftop PV and notified maximum demand are the two types of PV systems capacity boundaries considered in this study.