Open Access

The electrification of transportation requires the development of smart charging management systems for electric vehicles to optimize grid performance and enhance user satisfaction. However, existing methods often reduce multi-objective problems to single-objective formulations, limiting their ability to balance conflicting objectives and requiring iterative runs for diverse solutions. In this study, we propose a Multi-Objective Evolutionary Reinforcement Learning (MOEvoRL) framework to optimize electric vehicle charging strategies and discover multiple policies within a single training run. Our approach focuses on maximizing the batteries’ state of charge, increasing photovoltaic power consumption, reducing peak loads, and smoothing the overall load on the grid. Simultaneously, it adheres to essential grid constraints, such as load balancing and grid connection node limits, to ensure grid stability, efficiency, and real-world applicability. MOEvoRL utilizes the exploratory power of Evolutionary Algorithms and the sequential decision-making strengths of Reinforcement Learning. By employing neuroevolution, we optimize the weights and topologies of policy networks. Our approach employs the Non-dominated Sorting Genetic Algorithm II, Strength Pareto Evolutionary Algorithm 2, and a modified NeuroEvolution of Augmenting Topologies as optimizers and benchmarks their performance against the gradient-based Multi-Objective Deep Deterministic Policy Gradient (MODDPG) and a single-objective DDPG that simplifies multiple objectives into a single objective using a linear scalarization function. The results show that MOEvoRL approaches are superior to MODDPG in terms of generalization, robustness, constraint compliance, and multi-objective optimization capabilities. In contrast, DDPG exhibits poor and unstable performance. This positions MOEvoRL as a robust strategy for managing electric vehicle charging while optimizing local grid loads.

With the increasing capabilities of machine learning (ML) and other artificial intelligence (AI) methods comes a growing interest from many fields of application to employ these methods to increase automation of work tasks and improve the efficiency and effectiveness of operations. However, the systems will only see effective use if they are trusted by those responsible for the task itself. False predictions and flawed decisions can have detrimental effects, for example, on human life in medical applications or financial interest in industry. Therefore, it is reasonable for stakeholders of these systems to want to understand the reasoning of AI systems. Methods that can make this insight available to stakeholders have increasingly be summarized under the term explainable AI (XAI). While approaches exist towards making black-box models explainable, the use of inherently explainable models can be more straightforward and promising. One family of algorithms producing inherently explainable models are. Learning Classifier Systems (LCSs). Despite their name, they are a general rule-based ML (RBML) method and representatives for all major ML tasks have been proposed. To classify LCSs based on their mode of operation, this work introduces a new system that is more precise than the current stateof- the-art and based on descriptive ML terminology. While most researchers in the past have focused primarily on LCSs’ algorithmic aspects, this work adopts a distinct perspective by approaching them through the lens of optimization. It discusses LCSs with regards to typical tasks involved in creating an ML model and what specific elements have to be optimized and how this is typically done. Critically, the task of model selection is usually performed by some metaheuristic component and involves the subtasks of how many rules to use and where to place them. This work also proposes a template to assess use case–specific explainability requirements based on multiple stakeholders’ inputs and extensively demonstrates its usage in a real-world manufacturing setting. There, stakeholders indeed request XAI models over black-box approaches and, according to their answers, LCSs should be a good fit. Additionally, the results laid out what LCS models in that application should look like which is, however, not achievable with the major state-of-the-art LCSs. Therefore, a new LCS, called the Supervised Rule-based learning system (SupRB), is introduced in this work that is simpler than previous LCSs with clearer optimization objectives and models that can fulfil the stakeholders’ requirements. In extensive testing on real-world data, SupRB demonstrates its capabilities of producing small yet accurate models that outperform those of well-established methods. This work also investigates numerous possible extensions for each component of SupRB with a special focus on its optimizers and presents the findings of the multiple studies in a comprehensive manner based on descriptive statistics, visualizations of results, and rigorous statistical testing. Then various paths for future research and application of SupRB are laid out which can advance the field of XAI considerably.

Organic Computing systems intended to solve real-world problems are usually equipped with various kinds of sensors and actuators in order to be able to interact with their surrounding environment. As any kind of physical hardware component, such sensors and actuators will fail after a usually unknown amount of time. Besides the obvious task of identifying or predicting hardware failures, an Organic Computing system will furthermore be responsible to assess if it is still able to function after a component breaks, as well as to plan maintenance or repair actions, which will most likely involve human repair workers. Within this work, three different approaches on how to prioritize such maintenance actions within the scope of an Organic Computing system are presented and evaluated.

Maintenance of complex machinery is time and resource intensive. Therefore, decreasing maintenance cycles by employing Predictive Maintenance (PdM) is sought after by many manufacturers of machines and can be a valuable selling point. However, currently PdM is a hard to solve problem getting increasingly harder with the complexity of the maintained system. One challenge is to adequately prepare data for model training and analysis. In this paper, we propose the use of expert knowledge–based preprocessing techniques to extend the standard data science–workflow. We define complex multi-purpose machinery as an application domain and test our proposed techniques on real-world data generated by numerous machines deployed in the wild. We find that our techniques enable and enhance model training.

We present MAHF, a Rust framework for the modular construction and subsequent evaluation of evolutionary algorithms, but also any other metaheuristic framework, including non-population-based and constructive approaches. We achieve high modularity and flexibility by splitting algorithms into components with a uniform interface and allowing communication through a shared blackboard. Nevertheless, MAHF is aimed at being easy to use and adapt to the specific purposes of different practitioners. To this end, this paper focuses on providing a general description of the design of MAHF before illustrating its application with a variety of different use cases, ranging from simple extension of the set of implemented components and the subsequent construction of algorithms not present within the framework to hybridization approaches, which are often difficult to realize in specialized software frameworks. By providing these comprehensive examples, we aim to encourage others to utilize MAHF for their needs, evaluate its effectiveness, and improve upon its application.

Automated prediction systems based on machine learning (ML) are employed in practical applications with increasing frequency and stakeholders demand explanations of their decisions. ML algorithms that learn accurate sets of rules, such as learning classifier systems (LCSs), produce transparent and human-readable models by design. However, whether such models can be effectively used, both for predictions and analyses, strongly relies on the optimal placement and selection of rules (in ML this task is known as model selection). In this article, we broaden a previous analysis on a variety of techniques to efficiently place good rules within the search space based on their local prediction errors as well as their generality. This investigation is done within a specific pre-existing LCS, named SupRB, where the placement of rules and the selection of good subsets of rules are strictly separated—in contrast to other LCSs where these tasks sometimes blend. We compare two baselines, random search and -evolution strategy (ES), with six novelty search variants: three novelty-/fitness weighing variants and for each of those two differing approaches on the usage of the archiving mechanism. We find that random search is not sufficient and sensible criteria, i.e., error and generality, are indeed needed. However, we cannot confirm that the more complicated-to-explain novelty search variants would provide better results than -ES which allows a good balance between low error and low complexity in the resulting models.

Unlike in traditional Genetic Programming, Cartesian Genetic Programming (CGP) does not commonly feature a recombination/crossover operator, although recombination plays an important role in other evolutionary techniques, including Genetic Programming from which CGP originates. Instead, CGP mainly depends on mutation and selection operators in their evolutionary search. To this day, it is still unclear as to why CGP’s performance does not generally improve with the addition of crossover. In this work, we argue that CGP’s positional bias might be a reason for this phenomenon. This bias describes a skewed distribution of active and inactive nodes, which might lead to destructive behaviour of standard recombination operators. We provide a first assessment with preliminary results. No final conclusion to this hypothesis can be drawn yet, as more thorough evaluations must be done first. However, our first results show promising trends and may lay the foundationf or future work.

A large proportion of the energy consumed by private households is used for space heating and domestic hot water. In the context of the energy transition, the predominant aim is to reduce this consumption. In addition to implementing better energy standards in new buildings and refurbishing old buildings, intelligent energy management concepts can also contribute by operating heat generators according to demand based on an expected heat requirement. This requires forecasting models for heat demand to be as accurate and reliable as possible. In this paper, we present a case study of a newly built medium-sized living quarter in central Europe made up of 66 residential units from which we gathered consumption data for almost two years. Based on this data, we investigate the possibility of forecasting heat demand using a variety of time series models and offline and online machine learning (ML) techniques in a standard data science approach. We chose to analyze different modeling techniques as they can be used in different settings, where time series models require no additional data, offline ML needs a lot of data gathered up front, and online ML could be deployed from day one. A special focus lies on peak demand and outlier forecasting, as well as investigations into seasonal expert models. We also highlight the computational expense and explainability characteristics of the used models. We compare the used methods with naive models as well as each other, finding that time series models, as well as online ML, do not yield promising results. Accordingly, we will deploy one of the offline ML models in our real-world energy management system in the near future.