Application of Time Series Models to Analyze and Forecast the Influent Components of Wastewater Treatment Plants (WWTPs)

Time series models were developed to describe the statistical characteristics of the influent components of a wastewater treatment plant (WWTP) in Oak Ridge, TN. The data used to generate the models consisted of measurements of flow, temperature, BOD5, suspended solids, and ammonia nitrogen over nearly a 3-year period. The data set was characterized by periodically missing values during weekends and holidays. A two-directional exponential smoothing method was developed to estimate the values of those missing data points, prior to time series modeling. Several commonly used time series models, including the exponential smoothing model, ARIMA model, and the dynamic regression model, were applied to the time series of the five plant influent variables. The best models for each influent variable were selected based on various statistics and the ability of the models to forecast future values in the time series. The time series models were then used to simulate random time series of the influent variables with the same statistical characteristics as the original data. The original and randomly generated time series were characterized by similar means, standard deviations, cross-correlations and autocorrelation functions. These randomly generated time series can be used in conjunction with dynamic process models to evaluate the ability of a given design to effectively treat effluent flows under conditions of variability different than those present in the historical data.