This preview shows page 1. Sign up to view the full content.
Unformatted text preview: stem. Artificial neural networks, feedforward
networks with backpropagation, are then used to estimate the forecasting
model. [1, 39 – 41] 55 To conclude  except for the generation factors, all other factors are integrated
in the distribution substation forecasts. The distribution substation forecasts
provide important information to determine the area per sector load. With the
areabased approach and taking into account the spatial load growth
behaviour of the individual enduse supplies, the transmission substation load
saturation points can be determined more accurately. The load growth behaviour is defined in three different growth dynamics:
1) Dormant period: The time when no load growth occurs.
2) Growth ramp: During this period growth occurs at a relativelyrapid
rate, because of new construction in the area.
3) Saturated period: The area is “filled up”, fully developed.
This corresponds to the three distinct phases of Scurves. The maximum transmission system load is actually the summation of all those
different Scurves representing the different enduse profiles.
3.2.9 Sectors
The sector load forecasts are based on sector forecasts compiled by the
utility’s national key customer executives. For balancing the algorithm inputs,
only the major sectors are used. For example, mining, only the sector loads
for gold mines, platinum mines and coal mines are used. All other mine
activities are defined as “other mining”. The grouping of sectors is in accordance with the requirements for transmission investment criteria. 3.2.10 Area per Secto r Loads
The spatial load forecast divides the enduse load profiles into classes. Some
of the different classes are grouped together to be aligned with the
transmission investment criteria. That means that each distribution substation
load is subdivided into different loads according to the grouping of the classes.
All the individual grouped loads in a transmission substation supply area are
summated to determine the load mix as required for the investment criteria. 56 The loads are summated again to determine the area per sector loads (see
Chapter 5).
The area per sector loads, the area loads and the sector loads are then
checked for consensus. This is done as a matrix, where the area per sector
loads are the matrix elements, the column totals are the area loads and the
row totals are the sector loads. If the loads have been balanced, the column
totals will equal the area loads. The row totals are compared with the sector
load forecasts from the utility’s national key customer executives.
The componentbased approach (for load modelling) developed by EPRI
divides the load into classes such as residential, commercial, industrial,
agricultural, and mining. Each category of load class is represented in terms
of load components such as lighting, space heating, water heating, etc. Again
these enduse load profiles, divided into a number of groups (according to the
investment criteria), can be used for the componentbased modelling.
3.3 POINT LOADS The point loads (loads required for the load buses) are determined from the
balanced results (see Chapter 5). Referring to Figure 1.4.2, the point loads
are modelled and the results from the power flows are verified with the
expected loads. The export and import loads are also checked to ensure that
the modelled loads in the balancing algorithm are correct.
3.4 MAXIMUM AND MINIMUM LOADS The maximum and minimum loads for each area and transmission load are
estimated over the last ten years and then the balanced loads are scaled to
determine the expected load (see Chapter 5). 57 3.5 FACTORS CONSIDERED IN FORECASTS The next table summarises the factors that are integrated in the different
forecasts X X X Generation Pattern X X X X X X T RENDS NETWORKS X X Exports X E NVIRONMENTAL System Demand Power Station X TECHNOLOGICAL S ECTORS GENERATION LOAD PROFILES ASPECTS FORECASTS REGULATORY Table 3.5.1 – Summary of Factors X X
X Tx Losses
A rea X X X X X X Tx Substation X X X X X X X X Dx Substation X X X X X X X X X X X Sector
Area per Sector X X X X
X X X 3.6 MULTIPLE REGRESSION AND NEURAL NETWORKS Quantitative forecasting methods involve the analysis of historical data in an
attempt to predict future values of a variable (area loads) that is of interest.
Quantitative forecasting methods can be grouped into two kinds  time series
analysis and causal techniques.
The use of causal forecasting models (such as multiple regression) involves
the identification of other variables that are related to the variable to be
predicted. Once these related variables have been identified, a statistical model describing the relationship between these variables and the variable to
be forecast is developed. The statistical relationship derived is then used to
forecast the variable of interest. The area demands may be related to a 58 number of political and economic indicators, etc. In that case, the area demands would be referred to as the dependen...
View
Full
Document
This document was uploaded on 03/04/2014.
 Spring '14

Click to edit the document details