HomeRegional NewsEast AfricaIs your utility ready for the evolution of load demand forecasting?

Is your utility ready for the evolution of load demand forecasting?

By Eng. Mbae Ariel Mutegi, Chief Engineer for Network Audit at Kenya Power, Kenya

This article first appeared in ESI Africa Issue 5-2019.
Read the full digimag here or subscribe to receive a print copy here.

New, progressive regulations in the wholesale electricity market that encourage flexibility from market participants will go a long way to improve load forecast models. This will allow energy storage to participate in wholesale energy auctions as renewable energy uptake deepens, fundamentally shifting the electric power load demand dynamics.

Accurate short-term load forecasting is an important tool used by spot market players in the daily energy dispatch process in a power system. It is a critical ingredient for optimal generator unit commitment, economic dispatch, system security and stability assessment, contingency and ancillary services management, reserve setting, demand-side management, system maintenance and financial planning in power systems.

Load forecasting refers to the accurate prediction of the power demand within a given planning time frame. There are three load forecast time frames, with each forecast horizon playing a critical role in the overall planning of the business, as can be seen in Table 1.

Day-ahead load forecasting is a daunting task due to seasonality and dependence on a multiplicity of factors such as weather patterns, deregulation and unbundling of power systems, increased uptake of renewable resources, day of the week, public holiday, time of the month, and school holidays. An efficient load forecasting technique should employ the following principles: causality, reproducibility, functionality, sensitivity and simplicity. The models are generally divided into three categories; namely the statistical analysis models, artificial intelligence models, and Grey prediction models.

Statistical methods are based on a mathematical representation of the load curve. There are several statistical methods that have been widely used including regression techniques, time series methods, and exponential smoothing. As an example, the time series methods assume that future load demand only depends on historical demands. The accuracy of statistical models worsens during sudden changes in the factors that affect load demand. The main limitation of statistical modelling is that its forecasting accuracy is based on the availability of sufficient data samples, multiple complicated variables, and a number of statistical data assumptions.

Artificial intelligence (AI) forecasting models have shown an ability to give better performance in handling non-linearity and other challenges associated with statistical models. Their main advantage lies in the fact that they don’t need complex mathematical formulations or quantitative correlation between inputs and outputs. These methods include artificial neural networks, fuzzy logic, expert systems, genetic algorithm, and particle swarm optimisation. AI models have a number of limitations, key among them being that their accuracy is limited by the number of the training sample data.

The Grey model focuses on model uncertainty and lack of enough information in analysing and understanding systems through research on conditional analysis, prediction and decision-making. This model is suitable for application to system analysis, data processing, modelling, prediction, decision-making, and control. The model was developed to overcome the challenges associated with statistical analysis and AI forecasting models.

In recent years, there has been a lot of research work geared towards improvement of the Grey forecasting model. This has been mostly by combining the model with various statistical and AI methods. Hybrid load forecasting combining statistical and AI approaches has been explored with good results; e.g. the chaotic evolutionary algorithm that combines both the regression techniques and fuzzy logic.

An important aspect in load forecasting is seasonality that refers to regular, predictable and recurrent patterns in time series data. In the electricity industry, the main types of recurrent patterns include day of the week, month of the year, weekends, public and religious holidays, school holidays, festive season, climatic season (rainy, cold or hot), and festive season. Each seasonality has to be modelled accordingly so as to improve the forecasting accuracy.

Future outlook of load forecasting

In an electricity wholesale market, the power grid planning and technical support systems such as ancillary services and generation planning require very accurate load forecast data. Reasonable electricity prediction is key to guaranteeing the stability and efficiency of electricity supply. Complete digitisation of the energy sector will increase the availability of accurate data that will play a critical role in improving the seasonality accuracy of the forecasting models. Data analytics by way of mathematical modelling, statistical analysis, predictive modelling, predictive analysis, and machine learning to find meaningful and useful patterns from large data sets will go a long way toward improving load forecasting models.

This should inevitably result in an improvement in the energy production efficiency and lowering of the production, transmission and distribution costs. Analytics can help energy utilities to better understand peak demand management and offer incentives for consumers to reduce energy consumption by shedding unnecessary load. Global warming has, for example, complicated weather forecast accuracy, which directly impacts on the accuracy of load forecast models – thus accurate data analytics will help to work around the challenge.

Deeper penetration of renewable energy resources as more prosumers come on board, coupled with the intermittent nature of renewables, will further increase the challenge of load prediction accuracy. Emerging factors such as the use of aggregators to create virtual power plants and networks, feed-in tariffs and blockchain’s peer-to-peer electricity trade will further imply the need of a more versatile load forecast model.

A progressive shift from long-term power purchase agreements to electricity auction models further necessitates the development of robust load forecasting models. Auctions help a utility to match demand with supply by reducing cases of having stranded power, which force consumers to pay for idle plants through capacity charges. This implies cheaper, cleaner and more sustainable power investment projects. Globally, we have more than 70 countries that have embraced the auction model, including a number of African countries such as Uganda, South Africa, Ghana, Mauritius and Zambia.

Other subtle seasonality aspects to bear in mind are election cycles, intra-Africa trade, regional peace and conflict, economic growth, and the expansion of regional power pools – with the latter playing a big role in the accuracy of the forecast model. Close monitoring of such will be very important going forward.

In May 2016, the World Bank group through the African Renewable Energy Access Program (AFREA) released a report titled ‘Making power affordable for Africa and viable for its utilities’. One of the key highlights of this is that only 2 out of 39 power utilities in sub-Saharan Africa are profitable. Only Uganda and Seychelles were fully recovering their operational and capital costs. Thus we have a long way to go before Africa’s power utility business becomes fully modernized and economically viable. ESI


• Assessment of an adaptive load forecasting methodology in a smart grid demonstration project by Ricardo Vazquez et al in the Energies Journal, February 2017.

• The electricity load forecasting models: A critical systematic review by Corentin Kuster et al in the Sustainable Cities and Society Journal, August 2017.

• Artificial intelligence in short term electric load forecasting: A state of the art survey for the researcher by K. Metaxiotis et al in the Energy Conversion and Management magazine.

• Global experience of unbundling national power utilities by Michael Boulle, March 2019.

This article first appeared in ESI Africa Issue 5-2019.
Read the full digimag here or subscribe to receive a print copy here.

Guest Contributor
The views expressed in this article by the author are not necessarily those of the publishers and/or association partners. While every effort is made to ensure accuracy, the publisher and editors cannot be held responsible for any inaccurate information supplied and/or published.