Paolo Fazzini

In many multi-channel engineering problems, the information relevant to the task is concentrated in one channel, while the other information plays an auxiliary role. In forecasting, the quantity to be predicted is the primary source of information and the covariates support it; in image processing, the image itself is primary and contextual data are auxiliary. This paper introduces multi-channel neighborhood-constrained variational mode decomposition (NCVMD), an extension of variational mode decomposition (VMD) built on this hierarchy. Where multivariate VMD (MVMD) aligns the modes’ central frequencies symmetrically and exactly across all channels, NCVMD distinguishes a main channel from auxiliary ones and makes the degree of alignment tunable, reallocating the decomposition error away from the priority channel by a controllable amount. A second contribution is a formal analysis of the resulting optimization problem, which delimits what can be asserted and what cannot. Every subproblem of the alternating scheme is convex and admits a unique closed-form solution, and the main channel provably attains the unconstrained VMD optimum, which MVMD cannot. The joint problem, however, remains nonconvex, so global optimality is not guaranteed. Experiments on synthetic and real-world data confirm the predicted behaviour.

This paper introduces Neighborhood-Constrained Decomposition-based Prediction (NCD-Pred), the first system to leverage Neighborhood-Constrained Variational Mode Decomposition (NCVMD) for multichannel forecasting by integrating time series decomposition and neural networks. NCD-Pred leverages NCVMD to decompose a multichannel signal into simpler, band-limited components—referred to as intrinsic mode functions or simply modes—by prioritizing the most informative channel (the main channel) over less informative ones (the auxiliary channels) and bringing their central frequencies into alignment up to a tunable extent. This frequency synchronization provides a framework for cooperative mode forecasting, where predictions of signal components are recombined to produce the original signal prediction. For mode-level forecasting, Long Short-Term Memory (LSTM) networks are utilized. NCD-Pred’s performance is evaluated against similarly designed mode-level forecasting systems using a multichannel dataset with weak cross-correlation, representing power load on a large vessel. The results show that NCD-Pred outperforms benchmark methods, demonstrating its practical utility in real signal processing scenarios.

This paper introduces Multi-Channel Neighborhood-Constrained Variational Mode Decomposition (NCVMD), a novel extension of Variational Mode Decomposition (VMD). When working with multi-channel signals, aligning mode frequencies serves as a criterion for grouping modes with similar frequencies and enabling mode-level processing. This requirement must be achieved while maintaining an acceptable error in the decomposition, and has been investigated by an approach called Multivariate Variational Mode Decomposition (MVMD), which forces a perfect alignment of the channels' modes' central frequencies at the expense of the decomposition error. Our approach differs in two key ways: first, by making a hierarchical distinction in a multi-channel signal between a primary (main) channel and others that provide auxiliary information; and second, by allowing tunable alignment between channel decomposition. Both these aspects reduce the decomposition error in the main channel at the expense of the auxiliary channels, at a customizable degree. These distinctive features make the method particularly well-suited for applications such as weighted image fusion, signal denoising, and time series forecasting; in these domains, there are clear advantages to both establishing a hierarchical processing framework for the signal's channels and operating at mode level, where intrinsic oscillatory patterns can be extracted and processed independently. We demonstrate the effectiveness of NCVMD through experiments on both synthetic and real-world datasets. The results indicate that NCVMD preserves a pure VMD-based decomposition for the main channel, maintains acceptable decomposition errors for the auxiliary channels, and achieves a high degree of mode frequency alignment across channels.

This paper presents a novel hybrid forecasting approach that integrates time series decomposition with machine learning to enhance prediction accuracy. The proposed method employs Long Short-Term Memory (LSTM) networks to generate predictions from multivariate input time series, which are first decomposed using a newly developed algorithm derived from Variational Mode Decomposition (VMD), the Neighborhood Constrained Variational Mode Decomposition (NCVMD). NCVMD decomposes multivariate input time series into band-limited modes, prioritizing a main signal, the target forecast, over auxiliary signals and aligning their central frequencies within a controllable range. This alignment enables cooperative forecasting, wherein predictions from individual modes are integrated to form the final output. The proposed method is compared to a similar hybrid model based on traditional VMD. Experimental results obtained using multivariate input time series, where each variable represents a different component of a large passenger ship's power demand, demonstrate that the proposed NCVMD-based approach delivers superior forecasting performance, highlighting its effectiveness for complex signal prediction tasks.

This paper presents a Variational Mode Decomposition based (VMD) approach for multi-step prediction of univariate electrical load on large passenger ships. Specifically, VMD is adopted as a pre-processing step to enhance the performance of Long Short-Term Memory (LSTM)-based forecasting models. It is demonstrated that decomposing the signal into intrinsic mode functions via VMD before forecasting significantly outperforms (up to a 66% error reduction) the conventional approach of feeding the entire signal into the forecasting model. The resulting findings suggest that the integration of VMD into the forecasting pipeline could yield a useful tool for a more accurate and efficient power management on large passenger ships.

In this study, a comprehensive examination, both theoretically and practically, is undertaken on Multi-Agent Reinforcement Learning algorithms (MARL). The investigation is situated within the context of real adaptive traffic signal control (ATSC) scenarios, with the primary objective being to validate the algorithms theoretical framework and evaluate their effectiveness, robustness, and applicability in real-world settings. The study uses two traffic networks in the city of Bologna, Italy, as examples. Key findings underscore the necessity of situating the algorithms within the context of a Partially Observable Markov Decision Process (POMDP), inherently characterizing them as non-Markovian. The equations are reformulated within this framework. Simulation results reveal that one of the studied algorithms, MA2C, consistently achieves significant traffic de-congestion in the considered scenarios. In general, its performance continually improves over time, resulting in a reduction of running vehicles by a factor of approximately 70 at the conclusion of the simulation. A training strategy independent of the specific vehicle flow has been implemented, rendering it adaptable for use with various traffic loads.

Anomaly detection (AD) in modern electrical microgrids is critical for ensuring reliable and safe operation as well as in improving the operational and maintenance efficiency of the infrastructure. This is particularly true for on-board microgrids, given their inherent electrical weakness and predominantly isolated operation. This paper proposes a machine learning-based approach for performing anomaly detection (AD) in electrical power consumption of a large passenger ship. The developed AD method relies on a gate recurrent unit (GRU) autoencoder (AE) model trained and validated with a multivariate electrical power time series collected from a real-world vessel. The proposed GRU AE approach is compared with different AE models, training and testing all models on an unsupervised context, and their performance is evaluated using various metrics suitable for such an unsupervised context.

In this study, we present a statistical forecasting framework and assess its efficacy using a range of established machine learning algorithms for predicting Particulate Matter (PM) concentrations in the Arctic, specifically in Pallas (FI), Reykjavik (IS), and Tromso (NO). Our framework leverages historical ground measurements and 24 h predictions from nine models by the Copernicus Atmosphere Monitoring Service (CAMS) to provide PM₁₀ predictions for the following 24 h. Furthermore, we compare the performance of various memory cells based on artificial neural networks (ANN), including recurrent neural networks (RNNs), gated recurrent units (GRUs), long short-term memory networks (LSTMs), echo state networks (ESNs), and windowed multilayer perceptrons (MLPs). Regardless of the type of memory cell chosen, our results consistently show that the proposed framework outperforms the CAMS models in terms of mean squared error (MSE), with average improvements ranging from 25% to 40%. Furthermore, we examine the impact of outliers on the overall performance of the model.

In this study, we present a statistical forecasting framework and assess its efficacy using a range of established machine learning algorithms for predicting Particulate Matter (PM) concentrations in the Arctic, specifically in Pallas (FI), Reykjavik (IS), and Tromso (NO). Our framework leverages historical ground measures and 24-hour predictions from 9 models provided by the Copernicus Atmosphere Monitoring Service (CAMS) to provide PM predictions for the following 24 hours. Furthermore, we compare the performance of various memory cells based on artificial neural networks (ANN), including recurrent neural networks (RNNs), gated recurrent units (GRUs), long short-term memory networks (LSTMs), echo state networks (ESNs), and windowed multi-layer perceptrons (MLPs), which are commonly employed in time series forecasting tasks. Irrespective of the chosen memory cell type, our results demonstrate that the proposed framework consistently outperforms the CAM models in terms of mean squared error (MSE), with average improvements ranging from 25% to 40%. Additionally, we investigate the impact of outliers on the overall model performance.

Adaptive traffic signal control (ATSC) in urban traffic networks poses a challenging task due to the complicated dynamics arising in traffic systems. In recent years, several approaches based on multi-agent deep reinforcement learning (MARL) have been studied experimentally. These approaches propose distributed techniques in which each signalized intersection is seen as an agent in a stochastic game whose purpose is to optimize the flow of vehicles in its vicinity. In this setting, the systems evolves toward an equilibrium among the agents that shows beneficial for the whole traffic network. A recently developed multi-agent variant of the well-established advantage actor-critic (A2C) algorithm, called MA2C (multi-agent A2C) exploits the promising idea of some communication among the agents. In this view, the agents share their strategies with other neighbor agents, thereby stabilizing the learning process even when the agents grow in number and variety. We experimented MA2C in two traffic networks located in Bologna (Italy) and found that its action translates into a significant decrease of the amount of pollutants released into the environment.