[Paper Review] Short Term Load Forecasting Models in Czech Republic Using Soft Computing Paradigms

Muhammad Riaz Khan, Ajith Abraham|ArXiv.org|May 16, 2004

Energy Load and Power ForecastingEngineering8 references18 citations

TL;DR

This study evaluates six soft computing models—multilayer perceptron, Elman recurrent network, radial basis function network, Hopfield model, fuzzy inference system, and hybrid fuzzy neural network—for short-term electricity load forecasting in the Czech Republic using seven years of hourly load data. The hybrid fuzzy neural network and radial basis function networks outperformed others in predicting two-day-ahead electricity demand, demonstrating superior accuracy and robustness in complex, non-linear load patterns.

ABSTRACT

This paper presents a comparative study of six soft computing models namely multilayer perceptron networks, Elman recurrent neural network, radial basis function network, Hopfield model, fuzzy inference system and hybrid fuzzy neural network for the hourly electricity demand forecast of Czech Republic. The soft computing models were trained and tested using the actual hourly load data for seven years. A comparison of the proposed techniques is presented for predicting 2 day ahead demands for electricity. Simulation results indicate that hybrid fuzzy neural network and radial basis function networks are the best candidates for the analysis and forecasting of electricity demand.

Motivation & Objective

• To evaluate the performance of diverse soft computing models in forecasting short-term electricity demand in the Czech Republic.
• To identify the most accurate and reliable models for predicting two-day-ahead hourly electricity load.
• To compare traditional neural networks, recurrent networks, fuzzy systems, and hybrid models in handling non-linear, time-varying load data.
• To provide practical insights for power system operators on model selection for load forecasting.

Proposed method

• The study employs six soft computing paradigms: multilayer perceptron (MLP), Elman recurrent neural network (RNN), radial basis function (RBF) network, Hopfield network, fuzzy inference system (FIS), and hybrid fuzzy neural network (FNN).
• All models are trained and tested on seven years of actual hourly electricity load data from the Czech Republic.
• The forecasting task is set to predict 48 hours (two days) ahead, with performance evaluated using standard error metrics.
• The hybrid fuzzy neural network integrates fuzzy logic for rule-based reasoning with neural network learning for adaptive weight adjustment.
• The RBF network uses localized radial basis functions centered at data points to approximate complex non-linear load patterns.
• Model performance is compared using quantitative error metrics, with emphasis on mean absolute error and root mean square error.

Experimental results

Research questions

• RQ1 Which soft computing model provides the most accurate two-day-ahead short-term load forecasting for the Czech Republic?
• RQ2 How do hybrid models like fuzzy neural networks compare to standalone neural networks and fuzzy systems in load forecasting accuracy?
• RQ3 What is the relative performance of recurrent networks (e.g., Elman) versus feedforward networks (e.g., MLP, RBF) in capturing temporal load dynamics?
• RQ4 Can the Hopfield network effectively model non-linear load behavior despite its limitations in dynamic systems?
• RQ5 How do fuzzy inference systems perform in comparison to data-driven neural models in handling uncertainty and non-linearities in load data?

Key findings

• The hybrid fuzzy neural network achieved the lowest prediction error among all models, making it the most accurate for two-day-ahead load forecasting.
• The radial basis function (RBF) network demonstrated strong performance, ranking second in accuracy and showing high stability across test periods.
• Multilayer perceptron and Elman recurrent networks showed moderate accuracy but were outperformed by hybrid and RBF models.
• The Hopfield model exhibited poor forecasting performance, likely due to its inherent limitations in handling time-series and non-linear dynamics.
• The fuzzy inference system provided reasonable results but was less accurate than hybrid and RBF models, particularly in capturing rapid load changes.
• Overall, the hybrid fuzzy neural network and RBF networks were identified as the best-performing models for short-term load forecasting in the Czech Republic.