Development and validation of time series forecasting models for wastewater treatment applications

A wastewater treatment facility typically utilizes a combination of physical, chemical, and biological processes to achieve desired treatment goals. The interactions exist between treatment processes are numerous, complex, and therefore difficult to define and delineate in an explicit manner. However, statistical tools can be employed to directly establish the input-out correlation at a wastewater treatment facility without the needs of exploiting the complex interactions exist within the boundary of the facility. This “black box” approach can be advantageously employed to monitor, forecast, and control the treatment performance of the facility. Time-series forecasting models will be developed in this research as a means to ensure the full compliance of discharge requirements at wastewater treatment facilities. Two time-series models, univariate autoregressive integrated moving average (ARIMA) and multivariate Transfer Function models, will be developed and defined the input-output correlations that are imperative to the performance of a wastewater treatment facility. In addition to model development and validation, the forecasting capabilities of the time-series models developed will be tested using actual operational data collected at two different wastewater treatment facilities: a large municipal wastewater treatment facility and a small industrial wastewater treatment facility. The selection of these two vastly different facilities as target systems for model development and validation will be useful to determine the sensitivity and applicability of time-series models as forecasting tools. The protocols for data handling and model self-calibration will also be developed in this research to further facilitate and enhance the applications of time-series models in wastewater treatment