Forecasting models for maintenance work load with seasonal components

Abstract

This paper presents the results of applying time series forecasting techniques to the forecasting of maintenance work load. In particular, this paper discusses several models developed to forecast the electronics maintenance work load for a weather forecasting system located in Alaska. The maintenance work load for this system typically increases during the winter season. This is mostly due to the system's remote geographic locations and the additional travel time required to reach these locations during the harsh Alaskan winter. Several models were developed and evaluated on the basis of their data fit and forecasting accuracy of seasonal and non-seasonal electronics maintenance work load. In the first part of the analysis, a regression model that uses a serial autocorrelated error correction procedure was developed to model the non-seasonal components of the work load. Seasonal work load components were modeled using seasonal and cyclical indicator variables. The cyclical indicator variables were effective in modeling this system's seasonal work load behavior. A model that uses a combination of seasonal and cyclical indicator variables was also effective in this respect. In the second part of the analysis, seasonal autoregressive integrated moving average (ARIMA) techniques were used to model and forecast maintenance work load. A brief description of these forecasting methods and the procedures used to identify an optimal work load forecasting model are provided. Two seasonal ARIMA models were developed: The first model used only maintenance predictor variables; the second model used a combination of maintenance predictor variables and cyclical indicator variables. All of the models were evaluated on the basis of their goodness-of-fit and forecasting accuracy. A seasonal ARIMA model that uses a combination of maintenance predictor variables and cyclical indicator variables had the best goodness-of-fit and provided the most accurate maintenance work load forecast. Cyclical indicator variables were found to be extremely effective in modeling the seasonal behavior of the maintenance work load in both the causal and stochastic models.