Addressing Load Shedding in East Africa: The Potential Role of Machine Learning in Solving the Region’s Electricity Crisis
On August 30, 2024, the Kenyan power grid suffered a blackout of unusual proportions, plunging the capital Nairobi and six of the eight regions on the national grid into darkness. This phenomenon has only grown increasingly familiar in the East African country as it ended 2023 experiencing three nationwide blackouts in four months. Load shedding, the deliberate shutdown of electricity in parts of a power distribution network to prevent the collapse of the entire system, has become a daily challenge for many East Africans.
This issue is driven by multiple factors: aging infrastructure being unable to keep up with rapid urbanisation, financial constraints within the energy supply companies, and climate change-induced droughts diminishing the capabilities of hydroelectric power generation. The economic consequences are severe, with the World Bank estimating that power outages could cost countries in the region up to 5% of their total GDP. Machine learning-based energy demand forecasting and load profiling present solutions to reduce the impact of load shedding by enabling more accurate prediction and management of electricity demand.
Energy Demand Forecasting
In East Africa, energy demand forecasting has not been prioritised. The reasons for this include a lack of advanced technology that can be used for accurate demand forecasting alongside budgetary limitations that make it hard to allocate funds for it. Consequently, this leads to lack of high-quality historical data, making it challenging to conduct accurate forecasts. In this article, we shall dive into how machine learning (ML) can be used in making load forecasting efficient and more accurate to mitigate the negative effects of load shedding.
Energy demand forecasting is the process of predicting future energy demand based on the historical data provided. Energy suppliers use energy forecasting to predict the quantity that needs to be generated to ensure that there is efficient energy distribution. Incorporating the ever-changing consumer patterns into distribution is crucial, to ensure that each household and firm has access to sufficient energy to operate daily.
Load Profiling
Load profiling is the process of analysing energy consumption patterns of a specific market over time to understand when energy is being consumed. This method creates a detailed model of electricity usage patterns for various consumers, such as households and firms. By identifying high demand areas, utility companies can prioritise regions for infrastructure upgrades and targeted demand response programs. Effective demand forecasting allows utility service providers to optimise operations, enhance grid stability and improve customer satisfaction through efficient energy management.
A load profile is a graph that illustrates the variation in electricity usage over time derived from data collected at regular intervals. This data is analysed to identify consumption patterns, categorize different sectors, and understand demand variations with pricing models. The insights gained from these profiles are also used to forecast future energy consumption trends. The graph helps grid operators understand the consumption patterns within different sectors through highlighting the daily peak and off-peak consumption times. Short term load forecasting is the process of predicting the demand over a short time. These forecasts enable grid operators to make real-time decisions regarding the power distribution and avoid sudden outages. Moreover, the utility can optimise the usage of available resources in advance, to ensure that enough generation capacity is available to meet the demand.
How Load Profiling is Done in East Africa:
Smart and prepaid meters – The installation of meters has rapidly increased within households to be most common method in East Africa. These meters provide real-time data on electricity usage and can take readings in set time intervals, allowing utility companies to have a detailed load profile. Moreover, pre-paid metering is a system where the consumer purchases electricity tokens in advance, enabling energy suppliers to anticipate how much energy will be consumed by each household.
Manual meter reading – In areas where smart meters are not readily available, suppliers have utility workers that visit households and record consumption data. Despite its labour-intensive nature, it is still a widely used method in East Africa.
Data loggers – Some utility companies use data loggers that are installed at key points in the distribution network to monitor electricity flows – which correlates to its consumption.
What Does Electricity Demand Look Like in East Africa?
East Africa is filled with an extensive amount of untapped energy resources. However, only 50% of the total population has access to electricity services, with rural areas having the least exposure at around 30%. This is due to inadequate financial resources to connect the sparse population in rural areas to the national electric grid. Even where off-grid solar systems are prioritised in the rural areas, there are often inadequate resources available to cater to each household. Additionally, some of the products are too expensive for consumers, given their low disposable income. Nonetheless, East Africa’s energy supply has a lot of potential in the future due to its vast renewable energy reserves. With the prioritisation of increasing renewable energy supply, the power market in East Africa is predicted to grow at a compound annual growth rate (CAGR) of 3% between 2021 and 2026.
Exploring renewable energy sources more intentionally could hold the key to reducing load shedding as it diversifies the power supply portfolio. Small scale energy projects such as wind and solar are currently providing access to electricity in remote and rural areas
Leveraging Machine Learning for Energy Demand Forecasting
Energy demand forecasting can utilize statistical methods like a measure of the autoregressive integrated moving average (ARIMA) for short-term predictions based on past consumption patterns, while regression analysis incorporates various external factors such as temperature and time of day for more accurate results. Combining these traditional methods with machine learning (ML) techniques, like artificial neural networks (ANN) and support vector machines (SVM), enhances accuracy by modelling complex, non-linear relationships in energy consumption – making it possible to better anticipate and manage future demand.
The most effective method for short-term forecasting is likely an ARIMA prediction model, given that historical data shows strong patterns, due to its simplicity and interpretability. For incorporating external factors, regression analysis is preferable, as it can handle multiple input variables and easily extrapolate future demand. SVM models are suited for complex, non-linear relationships, but they require expert interpretation.
Anomaly Detection and Outlier Handling
Anomaly detection traditionally relied on manual data inspection by experts, but the increasing volume of data has made this method impractical. Consequently, automatic anomaly detection using machine learning techniques such as one class support vector machine (OCSVM), has become more prevalent. OCSVM identifies outliers by learning the distribution of a dataset, detecting data points that significantly deviate. This approach helps grid operators detect unusual energy consumption patterns, understand their causes, and prepare for disruptions, thus preventing unplanned power cuts due to load shedding.
Key factors leading to anomalies in energy consumption would be:
Social and cultural events – Major events like the football World Cup, public holidays, and music festivals can significantly boost energy consumption due to increased live streaming, decorations, and usage of lighting and sound systems.
Economic activities – Economic booms raise energy usage through heightened industrial activity, while recessions often reduce it. Production cycle changes also impact consumption, with lower usage during planning stages.
Environmental factors – Extreme weather, such as heatwaves or cold snaps, can cause unexpected spikes or drops in energy demand.
Machine Learning’s Large Potential and Pressing Challenges
The potential for the application of machine learning models to account for unique local factors is large. This nuanced approach to forecasting allows for more efficient resource allocation and better infrastructure planning, as it leverages region-specific insights to make accurate predictions. By effectively integrating diverse and localized data sources, ML models offer a significant advantage over traditional methods, ensuring that energy supply can be more reliably aligned with consumer demand.
Implementing machine learning models faces several challenges:
Limited access to comprehensive and accurate data can impair ML model performance.
Insufficient internet and computational resources hinder ML development and installation.
A lack of trained data scientists and experts complicates system development and maintenance.
High expenses for ML technology can be prohibitive for smaller organisations.
Frequent power outages and unreliable internet affect data collection and model reliability.
Conclusion
In conclusion, effective demand forecasting could play a crucial role in expanding and stabilising East Africa’s energy supply – ensuring reliable electricity access for both households and businesses. Where load shedding is common due to inadequate demand forecasting and lacklustre maintenance, techniques like load profiling and short-term load forecasting can enhance energy distribution. By accurately predicting consumption patterns and adjusting supply strategies, energy suppliers can mitigate these blackouts. Incorporating renewable energy sources, upgrading infrastructure, and using large-scale battery storage for excess energy can further stabilise supply. Through predictive analytics, technological innovation, and strategic planning, East African countries can align energy production with consumer demand – leading to more sustainable and resilient power systems for all.