Mohamed E. El-Hawary
PRELUDE
Since time immemorial, human communities have been preoccupied with foreseeing the future and events that lie ahead. In the past, shamans, soothsayers, oracles, and comets foretold the future on the basis of prevailing ideologies, customs, and past observations of past events. Present-day science of forecasting is based on skillfully blending statistical principles, ingenious deductions and observations and measurements of the causal interactions between social, economic, and physical quantities and of the underlying processes. For an insightful and highly readable historical treatment of the evolution of forecasting, Reference [1] is recommended.
The intent of this introductory chapter is to offer the reader a brief discussion of selected technologies and issues of forecasting in electric power systems, and specifically load and electricity forecasting contributions in Chapters 2â8.
FORECASTING: GENERAL CONSIDERATIONS
The term forecasting refers to the process of making statements about events whose actual outcomes (typically) have not yet been observed. Moreover, forecasting is a decision-making tool that deals with predicting future events, and the proper presentation and use of forecasts to help in budgeting, planning, and estimating future growth of a quantity (or quantities.) Prediction is a similar, but more general term. Both might refer to formal statistical methods. In the simplest terms, forecasting aids in predicting future outcomes based on past events and expert insights. It is generally accepted that while forecasts are rarely perfect, they are more accurate for grouped data than for individual items and for shorter than longer time periods. In hydrology, the terms âforecastâ and âforecastingâ are sometimes reserved for estimates of values at certain specific future times, while the term âpredictionâ is used for more general estimates, such as the number of times floods will occur over a long period.
There are diverse applications of forecasting, the most familiar applications are in the area of weather conditions, where temperature, precipitation, wind, barometric pressure are to be forecast. Typical applications of forecasting in the business domain include Supply Chain Management which makes sure that the necessary productive resources (capital, labor, component parts, and the like) are always available to manufacture the required output to meet consumer demand and from that estimate determine the necessary resources to produce the forecasted amount of output. In other words, forecasts make sure that the right product is at the right place at the right time. Another business application of forecasting is Inventory Control which aims to maximize profits, efficient inventory management is required so as not to tie up idle inventory unnecessarily. Alternatively, accurate forecasting will help retailers reduce excess inventory and therefore increase profit margins. Accurate forecasting will also help retailers to meet consumer demand.
It is important to note that forecasts may be conditional in the sense that, if policy A is adopted then X will take place. The field of forecasting relies on judgment, uses intuition and experience in addition to quantitative or statistical methods. Quantitative forecasting relies on identifying repeated patterns in data, so it may take some time to see the same pattern repeat more than once. Combining judgment and quantitative forecasting gets the best results. The most trustworthy forecasts combine both methods to support their strengths and counteract their weaknesses.
FORECASTING IN ELECTRIC POWER SYSTEMS
Forecasting of electric power system variables is vital for many operational and planning functions. Historically, forecasting power system load has been a dominant application in the electric utility business. In this regard, forecasting the demand for water and gas in a corresponding utility occupy the same prominent position in the utilities business.
The advent of power system competition and deregulation has introduced new requirements for forecasting additional quantities, with varying degrees of importance. Forecasting the electrical energy price in power markets is most relevant along with other applications such as:
- Energy price forecasting and bidding strategy in power system markets
- Day-ahead prediction of residual capacity of energy storage unit of microgrid in islanded state
- Reservoir inflow forecasting
- Flood forecasting
The introduction of renewable energy sources, has introduced new forecasting challenges such as:
- Wind power forecasting
- Photovoltaic and solar power forecasting
- Marine currents for tidal, wave, and river turbines
In scanning the literature on forecasting in electric power systems, we encounter new challenges such as:
- Electric power consumption forecast of life of energy sources
- Power quality prediction
Our conversation will focus on advanced load and price forecasting approaches.
LOAD FORECASTING IN ELECTRIC POWER SYSTEMS
Some may argue that electricity load forecasting has reached a state of maturity, but the advent of electricity markets and the progress in renewable energy sources have changed the nature of electricity production and consumption. In their now classic review paper, Gross and Galiana [2] defined electric power systems' short-term load forecasting (STLF) as dealing with prediction times of the order of minutes, hours, or possibly half hour up to 168 hours or a few weeks for the short-term problem. It is natural to recognize that from a mathematical forecasting (prediction) technique point of view, this qualification is not essential because the techniques apply equally to longer-term forecasts of months and even longer. Since the load forecasts play a crucial role in the composition of these prices, they have become vital for the electricity industry. In the pre-competitive era, the basic variable of interest in STLF has been, typically, the hourly integrated total system load. In a competitive market, the participants such as power producers, independent system operators, and power aggregators determine the âsamplingâ frequency and hence the prediction duration. The term load may mean peak daily system load, or system load values at pre-defined times of the day, or the hourly (or weekly) system energy, or individual bus loads or energy levels. Short-term forecasting is closer to operations, while the long-term function is closer to system planning applications.
The forecasting models vary because the factors affecting the prediction vary.
The active power generation of the system follows the active power load at all times. Whole units must be brought on line or taken out on an hourly basis, and prediction of load over such intervals is essential. Unit commitment and spinning reserve allocation need STLF based on 24 hour predictions. Security assessment relies on a priori knowledge of the expected values of bus loads from 15 minutes to a few hours to allow detecting vulnerable situations and taking corrective measures. Regular maintenance scheduling requires load forecasts of 1 or 2 weeks to maintain a predetermined reliability level.
The behavior of an electric power system load depends on factors such as time, weather, and small random disturbances reflecting the inherent statistical nature of the load because not every user is affected in the same way by the time and weather effects. Time factors include weekly periodicity and seasonal variations. Temperature, humidity, intensity of light, wind speed, precipitation, and cloud cover are reasonably important weather-related variables known to modify power consumption. Annual load growth or decline is a factor that used to be relatively easy to identify, reflecting primarily the growth of sales of new electric equipment relative to preceding years.
Other attributes can be suggested, such as the geographically distributed nature of the load and the possible decomposition into residential, commercial, municipal, and industrial type loads.
Predicting load involves developing a model describing its behavior based on possibly abstract rules discerned by experienced operators, or it may be a concrete mathematical model. The rules used by the operator in load forecasting may extrapolate past load behavior correlated with expected future weather, and are conceptually not different from a mathematical model.
Establishing mathematical load forecasting models involves modeling, identification, and performance analysis. The mathematical models hypothesize the structure of a model relating load to the effects influencing its behavior based on physical observations. The identification step determines the values of those free parameters of the model which result in the closest âfitâ of the load behavior generated by the model to the actual observed load behavior. The last step tests the validity of the model to forecast load. If the last step indicates that the hypothesis of the modelling step was inadequate, one returns to the model step for a modification of the model structure and a repetition of the next two steps is necessary.
ELECTRICITY PRICE FORECASTING IN ELECTRIC POWER SYSTEMS
In competitive electricity markets, participants, such as generators, power suppliers, investors, and traders, require accurate electricity price forecasts to maximize their profits. Forecasting loads and prices in electricity markets are mutually intertwined activities, and errors in load forecasting will propagate to price forecasting. Unlike load forecasting, electricity price forecasting is much more complex because of the unique characteristics, uncertainties in operation, as well as the bidding strategies of market participants. The main features that make it so specific include the nonstorability of power, which implies that prices depend strongly on the power demand. Electricity prices depend on fuel prices, generation unit operation costs, weather conditions, and probably the most theoretically significant factor, the balance between overall system supply and demand. Another characteristic is the seasonal behavior of the electricity price at different levels (daily, weekly, and ann...