Optimal wastewater treatment plant synthesis and design problem solution methodology

NOELIA ALASINO; MIGUEL MUSSATI; NICOLÁS SCENNA; PIO AGUIRRE

Lisboa

Conferencia; EngOpt2010-2nd International Conference on Engineering Optimization; 2010

Instituto de Engenharia Mecánica (IDMEC) of Instituto Superior Técnico

In a previous work, a superstructure model developed for process synthesis (plant configuration), design (equipment dimensions) and optimization of operation conditions of activated sludge wastewater treatment plants, in continuous operation, accounting for phosphorus and nitrogen removal, was developed. There, the model was solved using a multiple initial points strategy. The superstructure embeds a chain of up to seven reaction compartments in-series followed by a secondary settler, and allows for flow distribution of the main process streams, i.e. bypasses and recycles among reaction compartments, sludge recycles from the sedimentation zone to any reactor, and fresh feed distribution and external carbon source dosage along the reaction zone. Each compartment operates in aerobic, anoxic, or anaerobic conditions according to its aeration flow rate selected and the streams fed to it. The Activated Sludge Model No. 3 extended with the Bio-P module for computing biological phosphorus removal are used to model the reaction compartments. The performance criterion selected is the minimization of the Net Present Value that includes investment and operating costs, while verifying compliance with the effluent permitted limits. The problema characteristics allowed posing it as a NLP, specifically a nonlinear programming problem with discontinuous derivatives DNLP, as it results in a highly nonlinear system with non-smooth functions. However, the mathematical model is difficult to solve. The superstructure network is complex because it embeds a large chain of reactors to contemplate the most widely used configurations for combined N and P removal. It is well known how complex a network becomes if the number of reactors increases. A distinctive feature of the model is the possibility of flow distribution of the main process streams, which provides flexibility and allows searching for novel or more efficient process configurations. By the other hand, the model used to represent the process units are highly nonlinear and incorporated non-smooth functions. Several locally optimal solutions can be found depending on the initial values set. The present work aims to develop an automatic resolution strategy to address this problem. The potential alternatives to evaluate include a successive initialization method departing from simplified NLP models, the resolution of linear problems to provide good initial points for the larger (rigorous) NLP models, or a combination of both, among others.