Northern Virginia Electric Cooperative (NOVEC) is an electricity distributor servicing parts of six Northern Virginia counties. In order to provide power to their customers, NOVEC purchases power in two ways: long-term bulk purchases and as-needed spot purchases. Bulk purchases occur up to five years in advance and are sized to meet expected power demand during that time period. In the event bulk purchases are insufficient to meet demand, spot purchases provide the power to cover the difference. Temperature fluctuations, mainly during the summer months, are a significant contributor to increased power demand in excess of the bulk purchase amount.

In order to purchase an appropriate amount of power through bulk purchases, NOVEC has developed a forecasting model that forecasts future power purchases over a 30-year horizon. NOVEC makes bulk power purchases based on the first 5 years of the forecast.

Based on recent warming trends, NOVEC believes that the current model may no longer be the best available and that a new weather-normalization method may better reflect weather trends. Improving the accuracy of the forecast would limit the amount of power that NOVEC has to buy beyond the bulk amount, thus decreasing costs. NOVEC requests analytical support to develop a new weather-normalization forecasting model or to determine that the existing model is the best available.

The purpose of this project is to develop a new weather normalization methodology to improve NOVEC's forecasting model by more accurately modeling future power demand. The model will take into account historical data as inputs: customer and power purchase total by month starting from 1983, hourly weather data starting from 1963, and Moody's state, county, and Washington, D.C. metro economic data starting from the 1970s. The end product of the project is a forecast of power demand for the next 30 years and a forecasting model that will give NOVEC the ability to perform additional analysis.

Utilizing the data sources described above, a weather normalization model was developed accoding to the following methodology.

Three sources of data were used: Moody's economic data and forecast, NOVEC's historic power purchases and customer total since 1983, and temperature data collected from Dulles Airport since 1963. This data was evaluated in the Data Validation step using Excel. The data was re-formatted and minor gaps in the data were filled using linear interpolation. Also, Heating Degree Days (HDD) and Cooling Degree Days (CDD) were calculated using the historic temperature data; these variables are used as measures of the impact of temperature on power demand. Excel is then used to launch an R model which utilizes the data in Excel to forecast the power demand. The methodology utilized is as follows: the economic variables and customer total are fit using a linear regression to the historic power demand. Based on this relationship and the historical HDD and CDD, the base power load and the seasonal power load are determined. In order to forecast future power demand, the forecasted economic variables are utilized to forecast the customer base in the future. The customer base informs the size of the base power load based on the historical relationship. In order to determine the total power demand, HDD and CDD are forecasted to determine the seasonal power load. The seasonal power load and the base load are combined to form the forecasted monthly power load. Three different methods were utilized to developed different forecasts: combined linear, split linear, and customer ratio method.

Each of the three forecasting methods produced a different 30-year power demand forecast.

Based on the statistical analysis of the different forecasts, the split linear method produces the most accurate forecast. We recommend that NOVEC use the split linear model along with the Holt-Winters method for forecasting HDD and CDD and utilize the capabilities provided by the Excel and R models to supplement their current forecasting methods. Additional alternatives that can be studied using the capabilities provided by the models are varying the economic scenarios, varying the range of input years for temperature data and power demand, varying values for determining the HDD and CDD, and varying the economic variables used to determine the customer base.