From Excel to AI: Viettel Solutions Empowers the Power Sector with Reduced Errors and Enhanced Operational Efficiency

Electricity Load Forecasting: A Major Challenge with Data and Traditional Methods

Electricity load forecasting is considered one of the most complex tasks in the power industry. Even a slight deviation of a few percentage points can lead to two extreme scenarios: power shortages resulting in rolling blackouts, localized outages, and costly diesel generator usage, or power surpluses causing idle generators, increased fuel costs, and operational inefficiencies. Both situations place significant pressure on grid operators, especially during periods of high load volatility.

According to Power Development Plan VIII, renewable energy sources (excluding hydropower) are projected to account for 28%-36% of the energy mix by 2030. By 2050, this share is expected to reach 74%-75%. Traditional manual load forecasting methods and renewable energy output predictions are increasingly inadequate for grid operation requirements.

While many utilities have achieved relatively low forecasting errors, manual methods still have significant limitations. Forecasting primarily relies on historical data, simple models like linear regression or moving averages, and individual expertise, making it difficult to capture complex, non-linear demand fluctuations. Additionally, the growing volume of data from multiple sources is challenging to process comprehensively and in a timely manner.

Collecting and analyzing data for manual forecasts is time-consuming and labor-intensive. Engineers must manually integrate information from 30-minute load intervals, weather data, field reports, and technical data. This human-dependent process often leads to delays, reduces reliability due to manual data entry errors, and creates lags in grid management decision-making.

Manual methods, based on fixed assumptions, struggle to adapt to unexpected events like natural disasters, policy changes, or sudden consumption spikes. Human subjectivity and information gaps further contribute to forecasting inaccuracies.

The biggest bottleneck lies in data quality and infrastructure. Mr. Nguyễn Chí Linh shared insights from real-world projects: “Many companies lack a clear data strategy, fail to identify critical data, store data haphazardly without prioritization, and lack a governance framework to ensure data quality—the foundation for AI applications and accurate decision-making.”

This lack of a data strategy results in insufficient information for load forecasting and broader business decision-making.

With hourly load fluctuations, rapidly increasing renewable energy shares, and increasingly unpredictable weather, Excel-based, experience-driven forecasting is insufficient for today’s fast-changing weather and market dynamics. The power industry needs a data-driven, continuously updated, and multidimensional forecasting approach. Viettel Solutions addresses this by rebuilding data foundations for forecasting and AI applications.

Mr. Nguyễn Chí Linh emphasized, “AI starts with data. If AI is the brain, data is the bloodstream. No matter how intelligent the brain, it needs blood to function creatively and make accurate decisions.” For AI-powered forecasting, high-quality, diverse input data is as crucial as the model itself.

Unified Data and AI Elevate Load Forecasting Standards

Recognizing data bottlenecks, Viettel Solutions overhauled the forecasting process using a unified data platform. In a pilot project with Ho Chi Minh City Power Corporation (EVNHCMC), the AirData platform integrated real-time operational data—48-cycle loads, hourly weather, socioeconomic indicators, special event calendars, and regional consumption history—eliminating manual data consolidation.

On this standardized data foundation, Viettel Solutions trained advanced AI models like LSTM (Long Short-Term Memory), a specialized recurrent neural network designed for long-sequence data such as time series. Hybrid models were also deployed, continuously validated, and optimized based on real-world performance.

Results showed significant AI improvements. In June 2025, AI forecasting error dropped to 3.59%, a substantial reduction from the traditional method’s 5.54%. This nearly 2% error decrease is not just a technical achievement but a strategic advancement, enabling more accurate grid management and proactive, sustainable operations.

The LSTM-based AI model, combined with weather data, demonstrated high load forecasting accuracy and potential for automating the entire forecasting process, making it faster and more stable. AI reduces reliance on manual methods, minimizes human error, and enhances adaptability to sudden demand fluctuations.

With 35-45% accuracy improvements, Viettel Solutions recommends expanding AI to other power sector areas: renewable energy output forecasting using weather data, renewable energy pricing predictions for electricity markets, and predictive maintenance for power plants. These applications will optimize costs, enhance grid reliability, and support sustainable energy transitions.

Building a Data-AI Ecosystem for Digital Power Plants

Viettel Solutions views AI not as an endpoint but as a tool maximized by high-quality data and infrastructure. Mr. Nguyễn Chí Linh stated, “AI’s potential is fully realized only with accurate, complete, clean, and real-time data, supported by suitable infrastructure.” Viettel Solutions developed the Viettel Data Platform, a unified data hub integrating operational data from SCADA, DCS, and enterprise IT systems (ERP, inventory, maintenance schedules, and technical records).

Data standardization reduces report compilation time by 50-80% and provides clean, continuous data for weekly or monthly AI retraining. Digitization tools like OCR, IPA, and Voice AI convert technical documents, meeting minutes, and operational logs into structured data, enhancing the data repository and analysis quality.

With a robust data foundation, monitoring modules like APM Health (equipment health assessment) and APM Reliability (failure prediction) analyze machinery conditions and provide early risk alerts. Systems like Energy OMS (energy operation management) and VMS (plant operation monitoring) track generators, output, and incidents on a single interface, creating a data-AI ecosystem for proactive plant management, reducing reliance on individual expertise.

Viettel Solutions’ digital ecosystem not only improves forecasting accuracy but also transforms power sector operations from fragmented to seamless, data-driven, AI-accelerated models. Amid energy transitions and cost optimization pressures, this approach enables flexible, safe, and efficient grid operations.