Distributed learning for optimised operating of electrolyser modules

Abstract

The invention relates to a system (1) for improving the operation of electrolyser modules (2), the system (1) comprising a plurality of electrolysis systems (3) each having a plurality of electrolyser modules (2), and a control centre (4) which is connected to the electrolysis systems (3) for exchanging data and which is configured to collect model parameters of the electrolyser modules (2) of the electrolysis systems (3), to compress said model parameters and to update a model for electrolyser modules (2), wherein the electrolysis systems (3) each comprise one controller (5), connected to the electrolysis modules (2), and one local training module (6) which is connected to the controller (5) and is configured to update a statistical model of the electrolyser modules (2) from controller data, to forward model parameters of the updated statistical model to the control centre (4), to receive model parameters processed in the control centre (4) and to forward same, in prepared form, to the respective controller (5) for controlling the electrolyser modules (2). The invention also relates to a method for improving the operation of electrolyser modules (2) in electrolysis systems (3).

Description

Description JUEL Distributed learning for optimized operation of electrolyzer modules TECHNICAL AREA

[0001] The invention relates to a system for improving the operation of electrolyzer modules and a corresponding method. BACKGROUND

[0002] The operating mode of electrolyzer modules has a significant influence on the degradation behavior of components such as membranes. It has been shown that excessive switching accelerates the degradation of electrolyzer modules. A gentle operating mode can help remedy this. Several approaches for data analysis and control optimization are known from the state of the art.

[0003] Model Predictive Control (MPC), for example, is a method for optimizing the control of complex, usually multivariable processes. It is based on a discrete-time dynamic model of the system and uses real-time data to calculate optimal control signals. MPC can be used to optimize the operating parameters of electrolyzers and ensure high efficiency. However, this method requires considerable computing power.

[0004] Alternatively, by developing detailed physical models and simulations of electrolyzers, various operating conditions can be virtually tested and optimal settings determined. These models can be used to predict performance under different conditions and thus optimize control. However, many electrochemical processes and their relationship to operating mode are still unknown and are very difficult to physically model.

[0005] A solution could be a data-driven approach using static models or machine learning approaches. This involves identifying correlations between degradation characteristics and operating modes and developing data-driven models. Models can be developed to detect anomalies, predict when maintenance is required, or more generally monitor the operating status of the electrolyzer.

[0006] Modeling can therefore provide a number of competitive advantages, such as longer operating times, robust plant operation, or generally a deeper understanding of the extremely complex behavior patterns of electrochemistry in the cell.

[0007] Machine learning requires large amounts of data, which are usually collected centrally to create a dataset for training. However, exclusive access to the operational data required for machine learning presents a particular obstacle. For example, the operator of an electrolysis plant might consider information about its operating mode to be sensitive and not make the pure operational data available for further analysis.

[0008] To date, experimental investigations, physical, or data-driven modeling have been used to determine surrogate models that allow predictions of key performance indicators, such as degradation, and thus facilitate a more intelligent and sustainable operating strategy. However, the multitude of different environmental influences does not allow for tailored modeling for a specific plant with its associated environmental influences.

[0009] The object of the invention is to provide a system for improving the operation of electrolyzer modules. It is also an object of the invention to provide a method for improving the operation of electrolyzer modules in electrolysis plants.

[0010] The invention solves the problem directed to a system for improving the operation of electrolyzer modules by providing that such a system comprises a plurality of electrolysis plants, each with a plurality of electrolyzer modules, as well as a central unit connected to the electrolysis plants for the exchange of data, which central unit is configured to collect model parameters of the electrolyzer modules of the electrolysis plants, to condense these model parameters and to update a model for electrolyzer modules, wherein the electrolysis plants each comprise a controller connected to the electrolyzer modules and a local training module connected to the controller, which is configured to update a statistical model of the electrolyzer modules from controller data, forward model parameters of the updated statistical model to the central unit,receives model parameters processed in the control center and forwards them to the respective controller for controlling the electrolyzer modules.

[0011] In their article entitled "Communication-Efficient Learning of Deep Networks from Decentralized Data" https: / / arxiv.org / abs / 1602.05629, the authors H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Areas describe a method for using the wealth of data from modern mobile devices to create learning models for specific applications. These models can lead to an improved user experience, for example, through improved language models for speech recognition or easier text input. While mobile device hardware may vary, this is irrelevant for improving the same application. Since this extensive data is often privacy-sensitive, very large, or both, they propose leaving the training data on the respective mobile devices and creating a common model by aggregating locally computed updates.

[0012] Such an approach in machine learning, in which a distributed network of devices collaborates on a model without sharing central data, is called “federated learning.” This approach can be used in It can be useful for mobile phones because it improves data privacy and leverages the computing power of individual devices. However, there are several factors to consider for electrolysis systems that might prevent a professional from applying federated learning.

[0013] Electrolysis plants are typically large-scale industrial facilities for producing hydrogen by splitting water into oxygen and hydrogen. The technical requirements and processes of these plants differ significantly from those of mobile phones and other devices where federated learning is commonly used.

[0014] Furthermore, electrolysis systems are less widespread than mobile phones, which reduces the usefulness of federated learning, as this approach typically relies on a large number of devices.

[0015] For electrolysis plants, data protection may be less critical, and established mechanisms for data exchange between plant operators may already be in place.

[0016] Finally, unlike mobile phones, electrolysis systems may not have sufficient computing resources to run machine learning algorithms, which could affect their performance or require additional hardware.

[0017] Overall, a professional would therefore not necessarily consider federated learning to be suitable for electrolysis plants, especially with regard to challenges in implementing federated learning, such as coordinating model updates, ensuring data security, and dealing with devices or plants that have different computing power and network connections.

[0018] Nevertheless, federated learning also has advantages for electrolyzer modules. For example, compared to the above-mentioned established mechanisms for the Data exchange between plant operators is an improvement in terms of Data protection is achieved. Furthermore, the communication requirements are reduced because, instead of transmitting large amounts of data over the network, only model updates are transmitted, resulting in a lower network load. Regarding the common assumption that electrolysis plants, unlike mobile phones, may not have sufficient computing resources to run machine learning algorithms, which could impair their performance or require additional hardware, it has been shown that training the models on the local training modules can utilize their computing power instead of providing additional centralized computing resources.

[0019] In an advantageous embodiment of the invention, the local training modules are configured to learn relationships between operating modes and performance indicators of the electrolyzer modules.

[0020] By understanding the relationships between operating modes and key performance indicators, operators can optimize electrolyzer modules to achieve higher efficiency and better energy utilization. This can lead to reduced energy consumption and operating costs. Recognizing patterns and correlations enables targeted adjustment of the plant's operating parameters. This allows operators to ensure more stable and efficient production, which can extend the plant's service life and reduce maintenance costs.

[0021] Since electrolyzer modules can have different operating conditions, learning the interrelationships also helps adapt the system to changing environmental conditions or load requirements. This enables more flexible and efficient use of the electrolysis system.

[0022] Furthermore, potential problems or errors can be identified early on. This enables preventive maintenance to avoid failures and maximize plant availability.

[0023] Finally, learning these relationships contributes to the accumulation of knowledge about the optimal operating conditions and performance metrics of the electrolyzer modules. This knowledge can be used for future plant planning and optimization or shared between different plants to improve the efficiency and performance of the electrolysis industry as a whole.

[0024] In a further advantageous embodiment of the invention, the local training modules are configured to generate and / or update a statistical model for electrolyzer modules based on data representative of operating characteristics of the electrolyzer modules and to perform an optimization of the operation of electrolyzer modules based on this statistical model.

[0025] It is advisable for the optimization to include one or more of the following: overall efficiency of the electrolysis plants, degradation or service life of the electrolysis plants, degradation or service life of each of the electrolyzer modules, operational reliability, and maintenance patterns. This leads to a holistic improvement in plant performance. Advantageously, the local training modules can be synchronized by the central office. This ensures that all training modules have access to the same information and the latest model status. This contributes to the consistency and effectiveness of the learning process. Furthermore, the central office can better monitor the performance of the individual electrolysis plants and intervene as needed, thus identifying and resolving problems or inefficiencies at an early stage. Finally, synchronizing the local training modules simplifies the maintenance and updating of the plant model and ensures that all plants benefit from the latest findings and optimizations. [00271 It is also advantageous if the data is transmitted periodically to the control center. Through the periodic transmission of data, the control center is always informed about the current operating status of the electrolysis plants. This enables a timely response to potential problems, changes or inefficiencies, enabling faster adaptation and optimization.

[0028] Furthermore, regular data transmission allows the central office to better identify long-term trends and patterns in the electrolysis plants' operating data. This information can be used to identify potential for improvement or to predict future developments and requirements.

[0029] A simple and therefore advantageous way to summarize the collected data is arithmetic averaging.

[0030] The object directed to a method is achieved by a method for improving the operation of electrolyzer modules in electrolysis plants, wherein controllers in the electrolysis plants control and monitor the electrolysis modules, wherein local training modules of the electrolysis plants update a statistical model of the electrolyzer modules from controller data and forward the model parameters thus determined to a central location, wherein the central location collects these model parameters of the electrolyzer modules of the electrolysis plants, condenses them and updates a model for electrolyzer modules, wherein updated model parameters are distributed to the local training modules of the electrolysis plants and transmitted to the controllers in a processed form.

[0031] It is advantageous to learn about the relationships between operating modes and performance indicators of the electrolyzer modules.

[0032] It is advisable to generate and / or update a statistical model for electrolyzer modules in the electrolysis plants based on data representative of the operating characteristics of the electrolyzer modules and to optimize the operation of electrolyzer modules on the basis of this statistical model.

[0033] In particular, it is useful if the optimization includes one or more of the following points: overall efficiency of the electrolysis plants, Degradation or service life of the electrolysis plants, degradation or Service life of each of the electrolyzer modules, operational reliability and maintenance patterns.

[0034] It is advantageous if both the local training modules are synchronized by the central station and the data is periodically transferred from the local training modules to the central station.

[0035] It is also advantageous if the data are arithmetically averaged at the central station. The key features of the invention are the local training modules (local learners), which learn relationships between operating mode and KPIs (such as degradation) at the local operator plants. These local training modules are synchronized via a central server (master). The parameters of the local training modules are queried at regular intervals and transmitted to the central server. Since these are abstract model parameters, no conclusions can be drawn about the operating mode of the electrolysis plant operators. Sensitive information is therefore not transmitted. This is done for many different plants or operators. The collected model parameters are then aggregated on the master server. A very simple form, for example, is arithmetic averaging. The averaged model parameters are then transmitted back to the various plants.Through these updates, the predictive capability, robustness and generalizability of the local training modules based on the connected fleet can be significantly increased and dynamically adapted.

[0037] Dynamic, data-driven updates can be pushed to the plants and learned from data that is normally inaccessible. This allows the full potential of machine learning to be exploited and provides electrolysis plant operators with local predictive models based on data from the entire fleet, without actively using operator data. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] FIG System for improving the operation of electrolyzer modules according to the invention. DESCRIPTION OF THE EMBODIMENT

[0039] The figure shows the structure of a system 1 according to the invention for improving the operation of electrolyzer modules 2. In the embodiment of the figure, the system 1 comprises three electrolysis systems 3, each with three electrolyzer modules 2.

[0040] A central unit 4 is connected to the electrolysis plants 3 for data exchange. The central unit 4 is configured to collect 7 model parameters of the electrolyzer modules 2 of the electrolysis plants 3, condense 8 these model parameters, for example, perform arithmetic averaging, and update 9 a model for the electrolyzer modules 2.

[0041] The electrolysis systems 3 each comprise a controller 5 connected to the electrolyzer modules 2, which in the exemplary embodiment of the figure is in turn connected to a local training module 6 via a database 10. The local training module 6 is configured to update a statistical model of the electrolyzer modules 2 from data from the controller 5 and, in doing so, learn relationships between operating modes and key performance indicators of the electrolyzer modules, forward model parameters of the updated statistical model to the central unit 4, preferably periodically and in a synchronized manner (solid line), receive model parameters processed in the central unit 4 (dashed line), and, on the basis of these, forward corresponding instructions to the respective controller 5 for controlling the electrolyzer modules 2.

[0042] In particular, the local training modules 6 are configured to generate a statistical model for electrolyzer modules 2 based on data representative of operating characteristics of the electrolyzer modules 2 and / or update and optimize the operation of electrolyzer modules 2 based on this statistical model.

[0043] The optimization includes one or more of the following points: Overall efficiency of the electrolysis plants 3, degradation or lifetime of the electrolysis plants 3, degradation or lifetime of each of the electrolyzer modules 2, Operational safety and maintenance patterns. REFERENCE NUMBER LIST 1 system 2 Electrolyzer module 3 Electrolysis plant 4 Headquarters 5 controllers 6 local training module 7 Collecting model parameters 8 Condensing model parameters 9 Updating model parameters 10 Database

Claims

Claims What is claimed: 1 . System (1) for improving the operation of electrolyzer modules (2), the system (1) comprising a plurality of electrolysis systems (3), each with a plurality of electrolyzer modules (2), and a central unit (4) connected to the electrolysis systems (3) for data exchange, which central unit is configured to collect model parameters of the electrolyzer modules (2) of the electrolysis systems (3), to condense these model parameters, and to update a model for electrolyzer modules (2), wherein the electrolysis systems (3) each comprise a controller (5) connected to the electrolyzer modules (2) and a local training module (6) connected to the controller (5), which is configured to update a statistical model of the electrolyzer modules (2) from controller data, and to forward model parameters of the updated statistical model to the central unit (4),receives model parameters processed in the control center (4) and forwards them to the respective controller (5) for controlling the electrolyzer modules (2).

2. The system (1) for improving the operation of electrolyzer modules (2) according to claim 1, wherein the local training modules (6) are configured to learn relationships between operating modes and performance indicators of the electrolyzer modules (2).

3. The system (1) for improving the operation of electrolyzer modules (2) according to one of claims 1 or 2, wherein the local training modules (6) are configured to generate and / or update a statistical model for electrolyzer modules (2) based on data representative of operating characteristics of the electrolyzer modules (2) and to perform an optimization of the operation of electrolyzer modules (2) based on this statistical model.

4. The system (1) for improving the operation of electrolyzer modules (2) according to claim 3, wherein the optimization comprises one or more of the following points: overall efficiency of the electrolysis systems (3), degradation or lifetime of the electrolysis systems (3), degradation or lifetime of each of the electrolyzer modules (2), operational reliability and maintenance patterns.

5. The system (1) for improving the operation of electrolyzer modules (2) according to one of the preceding claims, wherein the local training modules (6) can be synchronized by the central unit (4).

6. The system (1) for improving the operation of electrolyzer modules (2) according to one of the preceding claims, wherein the system is configured such that the data is transmitted periodically to the central unit (4).

7. The system (1) for improving the operation of electrolyzer modules (2) according to one of the preceding claims, wherein the data are summarized by means of arithmetic averaging.

8. Method for improving the operation of electrolyzer modules (2) in electrolysis plants (3), wherein controllers (5) in the electrolysis plants (3) control and monitor the electrolysis modules (2), wherein local training modules (6) of the electrolysis plants (3) update a statistical model of the electrolyzer modules (2) from controller data and forward model parameters thus determined to a central station (4), wherein the central station (4) collects and condenses these model parameters of the electrolyzer modules (2) of the electrolysis plants (3) and updates a model for electrolyzer modules (2), wherein updated model parameters are distributed to the local training modules (6) of the electrolysis plants (3) and transmitted to the controllers (5) in a processed form.

9. Method according to claim 8, wherein relationships between operating modes and performance indicators of the electrolyzer modules (2) are learned in the electrolysis plants (3).

10. A process according to any one of claims 8 or 9, wherein in the electrolysis plants (3) a statistical model for electrolyzer modules (2) is generated and / or updated on the basis of data representative of operating characteristics of the electrolyzer modules (2), and an optimization of the operation of the electrolyzer modules (2) is carried out on the basis of this statistical model.

11. The method according to claim 10, wherein the optimization comprises one or more of the following points: overall efficiency of the electrolysis plants (3), degradation or lifetime of the electrolysis plants (3), degradation or lifetime of each of the electrolyzer modules (2), operational reliability and Maintenance patterns.

12. Method according to one of claims 8 to 11, wherein the local training modules (6) are synchronized by the central unit (4).

13. Method according to one of claims 8 to 12, wherein the data are transmitted periodically from the local training modules (6) to the central unit (4).

14. Method according to one of claims 8 to 13, wherein the data in the central (4) are arithmetically averaged.

Images