Forecasting with artificial neural networks: the state of the art

Introduction to linear and nonlinear programming

Load forecasting has become one of the major areas of research in electrical engineering, and most traditional forecasting models and artificial intelligence techniques have been tried out in this task. Artificial neural networks (NNs) have lately received much attention, and a great number of papers have reported successful experiments and practical tests with them. Nevertheless, some authors remain skeptical, and believe that the advantages of using NNs in forecasting have not been systematically proved yet. In order to investigate the reasons for such skepticism, this review examines a collection of papers (published between 1991 and 1999) that report the application of NNs to short-term load forecasting. Our aim is to help to clarify the issue, by critically evaluating the ways in which the NNs proposed in these papers were designed and tested.

Neural networks for short-term load forecasting: a review and evaluation

Energy is vital for sustainable development of any nation – be it social, economic or environment. In the past decade energy consumption has increased exponentially globally. Energy management is crucial for the future economic prosperity and environmental security. Energy is linked to industrial production, agricultural output, health, access to water, population, education, quality of life, etc. Energy demand management is required for proper allocation of the available resources. During the last decade several new techniques are being used for energy demand management to accurately predict the future energy needs. In this paper an attempt is made to review the various energy demand forecasting models. Traditional methods such as time series, regression, econometric, ARIMA as well as soft computing techniques such as fuzzy logic, genetic algorithm, and neural networks are being extensively used for demand side management. Support vector regression, ant colony and particle swarm optimization are new techniques being adopted for energy demand forecasting. Bottom up models such as MARKAL and LEAP are also being used at the national and regional level for energy demand management.

Energy models for demand forecasting—A review

Preface. Acknowledgments. 1 Complex Electricity Markets. 1.1 Liberalization. 1.2 The Marketplace. 1.2.1 Power Pools and Power Exchanges. 1.2.2 Nodal and Zonal Pricing. 1.2.3 Market Structure. 1.2.4 Traded Products. 1.3 Europe. 1.3.1 The England and Wales Electricity Market. 1.3.2 The Nordic Market. 1.3.3 Price Setting at Nord Pool. 1.3.4 Continental Europe 13. 1.4 North America. 1.4.1 PJM Interconnection. 1.4.2 California and the Electricity Crisis. 1.4.3 Alberta and Ontario. 1.5 Australia and New Zealand. 1.6 Summary. 1.7 Further Reading. 2 Stylized Facts of Electricity Loads and Prices. 2.1 Introduction. 2.2 Price Spikes. 2.2.1 Case Study: The June 1998 Cinergy Price Spike. 2.2.2 When Supply Meets Demand. 2.2.3 What is Causing the Spikes?. 2.2.4 The Definition. 2.3 Seasonality. 2.3.1 Measuring Serial Correlation. 2.3.2 Spectral Analysis and the Periodogram. 2.3.3 Case Study: Seasonal Behavior of Electricity Prices and Loads. 2.4 Seasonal Decomposition. 2.4.1 Differencing. 2.4.2 Mean or Median Week. 2.4.3 Moving Average Technique. 2.4.4 Annual Seasonality and Spectral Decomposition. 2.4.5 Rolling Volatility Technique. 2.4.6 Case Study: Rolling Volatility in Practice. 2.4.7 Wavelet Decomposition. 2.4.8 Case Study: Wavelet Filtering of Nord Pool Hourly System Prices. 2.5 Mean Reversion. 2.5.1 R/S Analysis. 2.5.2 Detrended Fluctuation Analysis. 2.5.3 Periodogram Regression. 2.5.4 Average Wavelet Coefficient. 2.5.5 Case Study: Anti-persistence of Electricity Prices. 2.6 Distributions of Electricity Prices. 2.6.1 Stable Distributions. 2.6.2 Hyperbolic Distributions. 2.6.3 Case Study: Distribution of EEX Spot Prices. 2.6.4 Further Empirical Evidence and Possible Applications. 2.7 Summary. 2.8 Further Reading. 3 Modeling and Forecasting Electricity Loads. 3.1 Introduction. 3.2 Factors Affecting Load Patterns. 3.2.1 Case Study: Dealing with Missing Values and Outliers. 3.2.2 Time Factors. 3.2.3 Weather Conditions. 3.2.4 Case Study: California Weather vs Load. 3.2.5 Other Factors. 3.3 Overview of Artificial Intelligence-Based Methods. 3.4 Statistical Methods. 3.4.1 Similar-Day Method. 3.4.2 Exponential Smoothing. 3.4.3 Regression Methods. 3.4.4 Autoregressive Model. 3.4.5 Autoregressive Moving Average Model. 3.4.6 ARMA Model Identification. 3.4.7 Case Study: Modeling Daily Loads in California. 3.4.8 Autoregressive Integrated Moving Average Model. 3.4.9 Time Series Models with Exogenous Variables. 3.4.10 Case Study: Modeling Daily Loads in California with Exogenous Variables. 3.5 Summary. 3.6 Further Reading. 4 Modeling and Forecasting Electricity Prices. 4.1 Introduction. 4.2 Overview of Modeling Approaches. 4.3 Statistical Methods and Price Forecasting. 4.3.1 Exogenous Factors. 4.3.2 Spike Preprocessing. 4.3.3 How to Assess the Quality of Price Forecasts. 4.3.4 ARMA-type Models. 4.3.5 Time Series Models with Exogenous Variables. 4.3.6 Autoregressive GARCH Models. 4.3.7 Case Study: Forecasting Hourly CalPX Spot Prices with Linear Models. 4.3.8 Case Study: Is Spike Preprocessing Advantageous?. 4.3.9 Regime-Switching Models. 4.3.10 Calibration of Regime-Switching Models. 4.3.11 Case Study: Forecasting Hourly CalPX Spot Prices with Regime-Switching Models. 4.3.12 Interval Forecasts. 4.4 Quantitative Models and Derivatives Valuation. 4.4.1 Jump-Diffusion Models. 4.4.2 Calibration of Jump-Diffusion Models. 4.4.3 Case Study: A Mean-Reverting Jump-Diffusion Model for Nord Pool Spot Prices. 4.4.4 Hybrid Models. 4.4.5 Case Study: Regime-Switching Models for Nord Pool Spot Prices. 4.4.6 Hedging and the Use of Derivatives. 4.4.7 Derivatives Pricing and the Market Price of Risk. 4.4.8 Case Study: Asian-Style Electricity Options. 4.5 Summary. 4.6 Further Reading. Bibliography. Index.

Modeling and Forecasting Electricity Loads and Prices: A Statistical Approach

A genetic algorithm (GA) is proposed to optimise train movements using appropriate coast control that can be integrated within automatic train operation (ATO) systems. The coast control output for a train changes with the interstation distances and gradient profiles, and the current operating conditions of the mass rapid transit (MRT) system, namely: (i) train schedules; (ii) expected passenger loads; and (iii) expected track voltages. The algorithm generates an optimum coast control based on evaluation of the punctuality, riding comfort and energy consumption. Before the train sets off to the designated station, a coast control table is generated that will be referenced by the train at runtime for deciding when to initiate coasting or resume motoring control. Each coast control table is encoded into variable length chromosomes with each gene representing the relative position between stations where coasting should be initiated or terminated. Each generation is evolved from mating of the paired equal-length chromosomes with possibilities of crossover, mutations, gene duplications and gene deletions. The key feature of this method is that it has a solid mathematical foundation. Effectively, the implementation provides good, credible and reasonably fast solutions for this variable dimensional and multiobjective optimisation problem. The algorithm has the potential for online implementation for producing a coast control lookup table for each interstation run before the train sets off. The results, although preliminary, suggest that the method is promising.

Optimising train movements through coast control using genetic algorithms

Area load frequency control using fuzzy gain scheduling of PI controllers

An Intelligent Tuned Harmony Search algorithm for optimisation

This paper describes the implementation and forecasting results of a hybrid fuzzy neural technique, which combines neural network modeling, and techniques from fuzzy logic and fuzzy set theory for electric load forecasting. The strengths of this powerful technique lie in its ability to forecast accurately on weekdays, as well as, on weekends, public holidays, and days before and after public holidays. Furthermore, use of fuzzy logic effectively handles the load variations due to special events. The fuzzy-neural network (FNN) has been extensively tested on actual data obtained from a power system for 24-hour ahead prediction based on forecast weather information. Very impressive results, with an average error of 0.62% on weekdays, 0.83% on Saturdays and 1.17% on Sundays and public holidays have been obtained. This approach avoids complex mathematical calculations and training on many years of data, and is simple to implement on a personal computer.

C.S. Chang

Papers

Optimising train movements through coast control using genetic algorithms

Area load frequency control using fuzzy gain scheduling of PI controllers

An Intelligent Tuned Harmony Search algorithm for optimisation

Demand forecasting using fuzzy neural computation, with special emphasis on weekend and public holiday forecasting

Parallel Neural Network-Fuzzy Expert System Strategy for Short-term Load Forecasting System Implementation and Performance Evaluation