Swarm intelligence techniques have been created to respond to theoretical and practical global optimization problems. This paper puts forward an enhanced version of the firefly algorithm that corrects the acknowledged drawbacks of the original method, by an explicit exploration mechanism and a chaotic local search strategy. The resulting augmented approach was theoretically tested on two sets of bound-constrained benchmark functions from the CEC suites and practically validated for automatically selecting the optimal dropout rate for the regularization of deep neural networks. Despite their successful applications in a wide spectrum of different fields, one important problem that deep learning algorithms face is overfitting. The traditional way of preventing overfitting is to apply regularization; the first option in this sense is the choice of an adequate value for the dropout parameter. In order to demonstrate its ability in finding an optimal dropout rate, the boosted version of the firefly algorithm has been validated for the deep learning subfield of convolutional neural networks, with respect to five standard benchmark datasets for image processing: MNIST, Fashion-MNIST, Semeion, USPS and CIFAR-10. The performance of the proposed approach in both types of experiments was compared with other recent state-of-the-art methods. To prove that there are significant improvements in results, statistical tests were conducted. Based on the experimental data, it can be concluded that the proposed algorithm clearly outperforms other approaches.

An effective energy oversight represents a major concern throughout the world, and the problem has become even more stringent recently. The prediction of energy load and consumption depends on various factors such as temperature, plugged load, etc. The machine learning and deep learning (DL) approaches developed in the last decade provide a very high level of accuracy for various types of applications, including time-series forecasting. Accordingly, the number of prediction models for this task is continuously growing. The current study does not only overview the most recent and relevant DL for energy supply and demand, but it also emphasizes the fact that not many recent methods use parameter tuning for enhancing the results. To fill the abovementioned gap, in the research conducted for the purpose of this manuscript, a canonical and straightforward long short-term memory (LSTM) DL model for electricity load is developed and tuned for multivariate time-series forecasting. One open dataset from Europe is used as a benchmark, and the performance of LSTM models for a one-step-ahead prediction is evaluated. Reported results can be used as a benchmark for hybrid LSTM-optimization approaches for multivariate energy time-series forecasting in power systems. The current work highlights that parameter tuning leads to better results when using metaheuristics for this purpose in all cases: while grid search achieves a coefficient of determination ($R^2$) of 0.9136, the metaheuristic that led to the worst result is still notably better with the corresponding score of 0.9515.

Any evolutionary technique for multimodal optimization must answer two crucial questions in order to guarantee some success on a given task: How to most unboundedly distinguish between the different attraction basins and how to most accurately safeguard the consequently discovered solutions. This paper thus aims to present a novel technique that integrates the conservation of the best successive local individuals (as in the species conserving genetic algorithm) with a topological subpopulations separation (as in the multinational genetic algorithm) instead of the common but problematic radius-triggered manner. A special treatment for offspring integration, a more rigorous control on the allowed number and uniqueness of the resulting seeds, and a more efficient fitness evaluations budget management further augment a previously suggested naïve combination of the two algorithms. Experiments have been performed on a series of benchmark test functions, including a problem from engineering design. Comparison is primarily conducted to show the significant performance difference to the naïve combination; also the related radius-dependent conserving algorithm is subsequently addressed. Additionally, three more multimodal evolutionary methods, being either conceptually close, competitive as radius-based strategies, or recent state-of-the-art are also taken into account. We detect a clear advantage of three of the six algorithms that, in the case of our method, probably comes from the proper topological separation into subpopulations according to the existing attraction basins, independent of their locations in the function landscape. Additionally, an investigation of the parameter independence of the method as compared to the radius-compelled algorithms is systematically accomplished.

As solar energy generation has become more and more important for the economies of numerous countries in the last couple of decades, it is highly important to build accurate models for forecasting the amount of green energy that will be produced. Numerous recurrent deep learning approaches, mainly based on long short-term memory (LSTM), are proposed for dealing with such problems, but the most accurate models may differ from one test case to another with respect to architecture and hyperparameters. In the current study, the use of an LSTM and a bidirectional LSTM (BiLSTM) is proposed for dealing with a data collection that, besides the time series values denoting the solar energy generation, also comprises corresponding information about the weather. The proposed research additionally endows the models with hyperparameter tuning by means of an enhanced version of a recently proposed metaheuristic, the reptile search algorithm (RSA). The output of the proposed tuned recurrent neural network models is compared to the ones of several other state-of-the-art metaheuristic optimization approaches that are applied for the same task, using the same experimental setup, and the obtained results indicate the proposed approach as the better alternative. Moreover, the best recurrent model achieved the best results with $R^2$ of 0.604, and a normalized MSE value of 0.014, which yields an improvement of around 13% over traditional machine learning models.

Monitoring heart electrical activity is an effective way of detecting existing and developing conditions. This is usually performed as a non-invasive test using a network of up to 12 sensors (electrodes) on the chest and limbs to create an electrocardiogram (ECG). By visually observing these readings, experienced professionals can make accurate diagnoses and, if needed, request further testing. However, the training and experience needed to make accurate diagnoses are significant. This work explores the potential of recurrent neural networks for anomaly detection in ECG readings. Furthermore, to attain the best possible performance for these networks, training parameters, and network architectures are optimized using a modified version of the well-established particle swarm optimization algorithm. The performance of the optimized models is compared to models created by other contemporary optimizers, and the results show significant potential for real-world applications. Further analyses are carried out on the best-performing models to determine feature importance.