Every morning, water suppliers need to define their pump schedules for the next 24 h for drinking water production. Plans must be designed in such a way that drinking water is always available and the amount of unused drinking water pumped into the network is reduced. Therefore, operators must accurately estimate the next day’s water consumption profile. In real-life applications with standard consumption profiles, some expert system or vector autoregressive models are used. Still, in recent years, significant improvements for time series prediction have been achieved through special deep learning algorithms called long short-term memory (LSTM) networks. This paper investigates the applicability of LSTM models for water demand prediction and optimal pump control and compares LSTMs against other methods currently used by water suppliers. It is shown that LSTMs outperform other methods since they can easily integrate additional information like the day of the week or national holidays. Furthermore, the online- and transfer-learning capabilities of the LSTMs are investigated. It is shown that LSTMs only need a couple of days of training data to achieve reasonable results. As the focus of the paper is on the real-world application of LSTMs, data from two different water distribution plants are used for benchmarking. Finally, it is shown that the LSTMs significantly outperform the system currently in operation.

In modern industrial plants, process units are strongly cross-linked with eachother, and disturbances occurring in one unit potentially become plant-wide. This can leadto a flood of alarms at the supervisory control and data acquisition system, hiding the originalfault causing the disturbance. Hence, one major aim in fault diagnosis is to backtrackthe disturbance propagation path of the disturbance and to localize the root cause of thefault. Since detecting correlation in the data is not sufficient to describe the direction of thepropagation path, cause-effect dependencies among process variables need to be detected.Process variables that show a strong causal impact on other variables in the process comeinto consideration as being the root cause. In this paper, different data-driven methods areproposed, compared and combined that can detect causal relationships in data while solelyrelying on process data. The information of causal dependencies is used for localization ofthe root cause of a fault. All proposed methods consist of a statistical part, which determineswhether the disturbance traveling from one process variable to a second is significant, and aquantitative part, which calculates the causal information the first process variable has aboutthe second. The methods are tested on simulated data from a chemical stirred-tank reactorand on a laboratory plant.

Modern water distribution networks are equipped with a large amount of sensors to monitor the drinking water quality. To detect anomalies, usually each sensor contains its own threshold, but machine-learning algorithms become an alternative to reduce the parametrization effort. Still, one reason why they are not used in practice is the geographical restricted data access. Data is stored at the plant, but data scientists needed for the data analysis are situated elsewhere. To overcome this challenge, this paper proposes a cloud-based event-detection and reporting platform, which provides a possibility to use machine learning algorithms. The plants measurements are cyclically transferred into a secure cloud service where they are downloaded and analyzed from the data scientist. Results are made available as reports

Climate change is leading to a general shortage of raw water availability combined with more pronounced seasonality and dry phases. The goal of the collaborative research project TwinOptPRO is to contribute to EU-wide climate neutrality in 2050 by the minimization of energy supply for water treatment and pumps in drinking water distribution systems. For that purpose, a digital platform that combines different forecasting models with simulation and optimization modules was developed. The aim is to ensure secure and compliant supply to customers in the future while maximizing the use of renewable energy and minimizing costs.

Zusammenfassung: Trinkwasserversorgungsnetze sind komplexe Systeme, welche zurzeit mit einer großen Anzahl von Wasserqualitätssensoren ausgestattet werden. Eine wichtige Aufgabe der Sensoren ist hierbei, die Betreiber des Wassernetzes über mögliche Kontaminationen im Netz zu informieren. Die Analyse der Sensorwerte erfolgt dazu normalerweise manuell über den Betreiber. Datenbasierte Methoden haben sich in der Praxis noch nicht bewährt, da diese unter anderem eine große Anzahl an falsch detektierten Alarmen auslösen, welche beispielsweise aufgrund betrieblicher Änderungen auftreten. In diesem Beitrag wird daher ein Verfahren basierend auf der dynamischen PCA entwickelt, welches die Anzahl an Fehlalarmen reduziert, gleichzeitig aber weiterhin sensitiv bei tatsächlichen Kontaminationen reagiert. Der entwickelte Ansatz wird anhand von Experimenten an einem Laborprozess evaluiert.

Industrial plants usually consist of different process units which are strongly cross-linked to each other. This leads to the point that a voluntary or involuntary change in one unit (e.g. changing some process control parameter or having a malfunctioning value) can lead to unexpected results in another process unit. Hence, knowing which are the causing and which are the effecting process variables is of great interest. Still, depending on the underlying process and the characteristics of the excitation signal, directed connectivities can or can not be detected. Therefore, in this paper several types of dynamic SISO systems and excitation signals are defined for which a directed connectivity from input to output signal should be detected and from output to input should not be detected. As a method for the detection of directed influences Spectral Granger Causality is used, which has been extended with a surrogatebased significance test. This test is used to define if a directed influence exists from one process variable to another.