Wastewater Treatment Plant Data for Nutrient Removal System

TL;DR

The Agtrup (BlueKolding) dataset addresses Phosphorus removal optimization in wastewater treatment by delivering high-frequency ($2$ minutes) SCADA time-series from the Agtrup WWTP, spanning August $2021$ to August $2023$. It combines chemical and biological phosphorus removal processes and provides a 23-signal input vector along with clear variable taxonomy (Control, Objective, Exogenous), enabling robust forecasting, digital-twin simulations, and reinforcement learning-based control. The paper documents data collection via Hubgrade SCADA, a comprehensive preprocessing pipeline with LOCF and quality handling, and feature selection using Pearson correlation to pinpoint key drivers of $PO_4$ and $NH_4$ dynamics. This dataset thus supports data-driven optimization, predictive analytics, and ML-driven control strategies to enhance phosphorus removal efficiency and operational sustainability in wastewater management.

Abstract

This paper introduces the Agtrup (BlueKolding) dataset, collected from Denmark's Agtrup wastewater treatment plant, specifically designed to enhance phosphorus removal via chemical and biological methods. This rich dataset is assembled through a high-frequency Supervisory Control and Data Acquisition (SCADA) system data collection process, which captures a wide range of variables related to the operational dynamics of nutrient removal. It comprises time-series data featuring measurements sampled to a frequency of two minutes across various control, process, and environmental variables. The comprehensive dataset aims to foster significant advancements in wastewater management by supporting the development of sophisticated predictive models and optimizing operational strategies. By providing detailed insights into the interactions and efficiencies of chemical and biological phosphorus removal processes, the dataset serves as a vital resource for environmental researchers and engineers focused on improving the sustainability and effectiveness of wastewater treatment operations. The ultimate goal of this dataset is to facilitate the creation of digital twins and the application of machine learning techniques, such as deep reinforcement learning, to predict and enhance system performance under varying operational conditions.

Figures (2)

• Figure 1: Schematic of the phosphorus removal process in the plant with the flow lines: The iron salt is added to the inflow to the biological tanks, where P is removed, and a sensor in Tank 1 measures phosphate. A dosage of polyaluminum chloride is taken before the secondary settler to remove the remaining P, and the final phosphate concentration is measured at the outlet mohammadi2024deep.
• Figure 2: The dynamic changes of the different variables in the biological process of wastewater treatment data