SCCPD – Load Curve and Demand Forecasting System
Between 2001 and 2002, UniSoma developed for Companhia Paulista de Força e Luz (CPFL), in partnership with Serviços Especializados (SEST) and RAM Computer Systems, SCCPD – Load Curve and Demand Forecasting System. This system collects and stores information regarding the active and reactive load curves at substations, transformers and large CPFL consumers. Besides this, the system supports several types of analysis, such as, demand forecast, according to the ONS (National Electric System Operator) requirements.
A load curve is a graph that shows the amount of energy being consumed over time at a determined point of the electric network. It’s a fundamental piece of information for the correct sizing of equipments. In case of industrial plants, the load curve is specially monitored during peak hours, when the energy cost is highest – in some situations, the production is reduced at this moment, or temporarily ceased.
At companies that market energy, such as CPFL, load curves are used to measure distribution losses. The load curves are also used in elaboration of demand forecast. Such studies are conducted to calculate future tension levels, which provide the optimum resource planning allocation for operations and expansion of the distribution network. Besides this, the demand forecast of various national electric companies are consolidated by ONS to analyze the current energy market supply conditions of SIN (National Interconnected System), as well as the generation of the Annual Electrical Operations Plan. In the “Load Forecast Consolidation” document, ONS attributes responsibilities and establishes the systematic of the consolidation process.
In July 2001, in partnership with SEST (Specialized Services) and RAM Computer Systems, UniSoma began development of SCCPD – Load Curve and Demand Forecasting System. The main objectives set for the project, at the time, were:
- Supply a methodological support of demand forecast, at each electrical bus, for operations and transmission expansion planning studies at CPFL and ONS;
- Build a support tool for analyzing data from load curves at substations, generators and consumers.
From a transactional point of view, the System should consolidate into a single data the base daily load curves for equipments and CPFL consumers (a) stored in mainframe – 5 million daily load curves, collected from 1992 to 1997 and (b) measured by registers and converted into load curve data base according formatting processes.
In terms of demand forecasting, the greatest challenge identified from the beginning was the size of the problem. The ONS specifications for load forecasting at the time were as follows:
- Short Term Forecasting (Monthly and Quarterly)
- Global active load demand curves for one weekday, Saturday and Sunday, monthly, in MW.
- For each electrical bus, active and reactive load demand curves for one weekday, Saturday and Sunday, monthly, in MW and MVAr.
- Medium Term Forecasting (16 months)
- Global active load demand curves for one typical weekday and Saturdays, monthly, in MW.
- For each electrical bus, active and reactive load demand curves for one weekday and Saturday, monthly, in MW and MVAr.
- Long Term Forecasting (48 months)
- Global active load demand curves for one typical weekday and Saturday, monthly, in MW.
- For each electrical bus, active and reactive load demand curves for one typical weekday and Saturday, monthly, in MW and MVAr.
A simple exercise can illustrate the magnitude of the problem. Considering a population of 250 substations, a load curve with 96 points (every 15 minutes), 2 kinds of load curves (active and reactive), 3 types of days (Saturday, Sunday and a typical weekday) for short term forecasting and 2 types (Saturday and a typical weekday) for medium and long term forecasting, 4,6 million points are necessary for the forecasting.
SCCPD was delivered in mid 2008. The system is based on three layers. The user layer is composed of user interface components. From the business layer, logical drives control the sequencing and execution of business rules, as well as the transactional integrity of conducted operations. Finally, there is a data layer whose methods guarantee the integrity and availability of corporate information.
In practical terms, the Load Curve System (developed in Borland Delphi© on top of an Oracle© database) is basically composed of the following functionalities:
- Load Curve formatting – processes that transform the field data in to daily data records divideby active and reactive;
- Load Curve Visualizer – interface that allows the selection of load curves, graph visualizer and advanced functionalities such as zoom and 3D curves;
- Load Curve Operations – electronic calculator that can perform algebraic operations on the curves; it is an additional component of the visualizer;
- Typical Curve Calculation – procedure that selects, from a set of daily curves from a given month and substation or consumer, that which “best” represents the typical behavior;
- Monthly Balance – report generated in batch containing a series of monthly information regarding, for example, monthly energy, maximum demand load, date and time, coincident demand with maximum CPFL system demand, date and time of coincident demand, standard deviation,etc.
- Maximum and Minimum Demands – report that show the variation range of demands measured at substations throughout the month (maximum and minimum values), subdividing in weekdays, Saturdays and Sundays, with active load (KW) and apparent load (KVA).
For ONS demand forecasting an orthodox formulation of Holt Winter’s adaptive time series with additional trends and seasonality was opted. The result of the data treatment is a load curve, which represents well the daily behavior of substations for each month considering the calibration data. For calibration, a set of interest point were chosen (instead of the 96 points that compose the whole curve), afterwards applying a “mask” representing the hourly evolution profile. The methodology was implemented in PL/SQL.
Besides forecasting for ONS a “Ten Year Diverse Study” was conceived, regarding the forecasting for substations and for a 15 year horizon, the maximum demand stratified by demand class. This study guides CPFL planning giving information regarding the load forecast for substations for the next years. The methodology adopted for the “Diverse Study” was based on a series of hypothesis and procedures internally developed at the company and that constitute, broadly, in the (a) breakdown of a reference curve according to known types of consumer classes and (b) application of growth rates to disaggregated load curves. The disaggregation procedure could be represented in the form of a Linear Programming Model. Due to this, the application was developed using the AIMMS tool.
|AIMMS Service Partner UniSoma|