The optimal automatic start-up of anaerobic digesters has remained an elusive problem over the years to be solved at the lowest possible costs, including that of process monitoring. In this work, a non-linear model predictive control (NMPC) system was developed, under two proposed configurations, for the optimal start-up of anaerobic digesters treating soluble non-recalcitrant substrates. The minimum set of low cost practical control variables (CVs) selected for process start-up include (i) the effluent quality as acetate COD, (ii) the level of aceticlastic methanogenic biomass in the reactor, and (iii) the methane production rate (only for one of the NMPC configurations). The manipulated variables (MVs) consist of the volumetric inflow rates of the organic substrate, dilution water, and of a possible concentrated alkali addition. To be able to apply the above selected CVs (technically and economically feasible to measure/estimate), a simplified tailored AD model was specifically designed as the prediction model, integral part of the NMPC system. The NMPC system developed was evaluated for a case scenario consisting of the automatic start-up of a high rate AD reactor treating a readily biodegradable carbohydrate based substrate. The AD plant was virtually represented by the complex Anaerobic Digestion Model No. 1. Compared to other manual start-up strategies, the two configurations of the NMPC developed appeared to reach the target methane production rate faster (39 and 18 days for the NMPC versus 70-75 days for the manual strategies) together with an overall superior CV set-point tracking error performance. Interestingly, the two configurations of the NMPC developed appear to propose two very different, almost opposite, start-up feeding strategies to both eventually start-up the reactor successfully with no process destabilizations throughout. A number of practical scenarios were also considered to evaluate the NMPC configurations for robustness and any possible improvements. These tests indicate that the NMPC objective function formulation is a key factor of the success and robustness exhibited during start-up.

中文翻译：

---

厌氧消化器实际自动启动的模型预测最优控制系统。

多年来，最佳的厌氧消化器自动启动仍然是一个难以解决的问题，需要以最低的成本（包括过程监控）解决。在这项工作中，开发了一种非线性模型预测控制（NMPC）系统，该系统采用两种建议的配置，用于优化厌氧消化池的启动时间，该消化池用于处理可溶性非难降解性底物。选择用于过程启动的最低限度的低成本实用控制变量（CV），包括：（i）乙酸盐化学需氧量的出水水质；（ii）反应器中乙腈产甲烷生物质的水平；以及（iii）甲烷产生速率（仅适用于NMPC配置之一）。受控变量（MVs）由有机底物，稀释水的体积流入速率和可能的浓碱添加量组成。为了能够应用上面选择的CV（在技术上和经济上可以测量/估计），特别设计了简化的量身定制的AD模型作为预测模型，将其作为NMPC系统的组成部分。针对以下情况对开发的NMPC系统进行了评估：自动启动高速率AD反应器，该反应器处理易于生物降解的基于碳水化合物的底物。AD工厂实际上由1号复杂的厌氧消化模型代表。与其他手动启动策略相比，开发的NMPC的两种配置似乎可以更快地达到目标甲烷生产率（NMPC分别为39天和18天手动策略需要70-75天），以及总体上出色的CV设定点跟踪错误性能。有趣的是 所开发的NMPC的两种配置似乎提出了两种非常不同的，几乎相反的启动进料策略，以最终成功启动反应器，而不会造成整个过程的不稳定。还考虑了许多实际方案来评估NMPC配置的健壮性和任何可能的改进。这些测试表明，NMPC目标函数公式是启动过程中成功和强大的关键因素。