Problem statement:
Power utilities are responsible for purchasing and selling energy in the market to meet the demands of their customers. Accurate load forecasting is crucial, as it enables utility companies to provide electricity at the lowest possible prices for their customers. Overestimating energy needs can result in financial losses and potential penalties from state regulators due to the unused energy. Conversely, underestimating energy needs can force companies to purchase additional energy in 'real time' at higher prices. Most of the forecasted load is purchased 16 hours before the start of the "flow date" in the "day-ahead" energy market. The flow date refers to the day on which the energy is consumed by customers. Any remaining load is purchased in the "real-time" energy market to accommodate fluctuations in demand. For instance, on Monday morning, the day-ahead market opens at 6 AM and closes at 8 AM, during which a utility company buys the energy required to meet customer demand on Tuesday (starting at midnight). This means the company purchases energy approximately 16 hours before the flow date begins. A 16-hour load forecast is generated every morning, and typically, utility companies aim for a Mean Absolute Percentage Error (MAPE) of under 2% for these forecasts. Accurate load forecasting helps utility companies optimize their energy purchases and minimize costs, ultimately benefiting their customers and ensuring efficient operation within the energy market.
Task: While the accuracy of the 16-hour forecast is of paramount importance, power utilities also generate forecasts for various time horizons, such as 7-day, 90-day, and 1-year ahead forecasts. Long-term one-year electricity load forecasts are essential for several reasons, including efficient resource allocation, infrastructure planning, and integration of renewable energy sources. The objectives of this assignment are to create long-term, one-year forecasts for the following goals:
Goal 0: fitted values for the models used to achieve Goals 1 - 3 described below.
Goal 1: Forecast hourly load for 2006. Your team's ranking in the leaderboard for hourly load forecasting will be determined based on the Mean Absolute Percentage Error (MAPE) of your forecasts for the year 2006. A lower MAPE corresponds to a higher ranking.
Goal 2: Forecast daily peak loads for each day of the year 2006. Your team's ranking in the leaderboard for daily peak load forecasting will be determined based on the MAPE of your forecasts for the year 2006. A lower MAPE corresponds to a higher ranking.
Goal 3: Forecast the timing (hour) of the daily peak loads for the year 2006. Your team's ranking in the leaderboard for daily peak load timing forecasting will be determined based on the Mean Absolute Error (MAE) of your forecasts for the year 2006. A lower MAE corresponds to a higher ranking.
By participating in these goals, you will contribute to the development of more accurate long-term energy forecasts, which can ultimately benefit utility companies, customers, and the overall stability of power grids. Accurate one-year forecasts enable power utilities to make informed decisions on resource allocation, plan for necessary infrastructure upgrades, and better integrate renewable energy sources to meet the growing demand for clean energy.
Data: A US-based utility company has provided the dataset for this project. Load forecasting is heavily influenced by weather conditions, as they directly impact electricity consumption patterns (e.g., increased or decreased use of air conditioning or heating systems). The file CompetitionData.xlsx https://github.com/robertasgabrys/Forecasting contains not only hourly electricity load data but also hourly average, median, maximum, and minimum temperature (in Fahrenheit) for an undisclosed city in the US. The dataset includes electricity load and temperature data from 2002 to 2005 (a total of four years), while only temperature data is provided for 2006.
Here is a snapshot of the first day in 2002 data:
· During the first hour on January 1, 2002, the average, median, maximum, and minimum temperatures were recorded as 43, 43, 60, and 31 degrees Fahrenheit, respectively. In this same hour, a total of 1,384,494 MWh (megawatt-hours) of electricity load was consumed.
Deliverables: Please submit the following items for the competition:
· Competition Template File: Submit the completed SubmissionTemplate.xlsx file, which contains three sheets with your predictions:
o Goal 0: provide the hourly load fitted values for 2002-2005.
o Goal 1: In the sheet named "Goal 1", provide the hourly load forecast for 2006.
o Goal 2: In the sheet named "Goal 2", provide the daily peak load forecast (maximum hourly load for each day).
o Goal 3: In the sheet named "Goal 3", provide the forecast for the timing of the daily peak load (hour in which the peak load occurs).
The report should cover the models you built and their performance. In your report, ensure that you clearly explain the models you have developed, their underlying methodologies, and the overall performance of these models. Highlight any challenges you faced during the process and the solutions you employed to overcome them. Additionally, discuss any insights or patterns you identified in the data, and how they influenced your model selection and forecasting strategy.