Periodic Reporting for period 1 - EVOLVE (Electric Vehicles Point Location Optimisation via Vehicular Communications)
Periodo di rendicontazione: 2023-01-01 al 2024-12-31
The transportation sector is one of the leading contributors to greenhouse gas (GHG) emissions. According to the European Environmental Agency in 2021, 22% of GHG emissions were contributed by the transportation sector, which is the second-largest source of emissions. To meet the European Union set target of 60% lower emissions from transport by 2050 as compared to 1990, full-spectrum decarbonization is required. The transition towards low-carbon transport, i.e. electric vehicles (EVs) is vital for achieving these ambitious targets. The scope of EV charging management systems is multifaceted and multi-dimensional as it brings together various thematic disciplines including power management in smart grids, control systems, optimization, communication technologies, and data analytics with an underlying common objective to design and develop strategies to boost EV performance, enhance EV driver's QoE, provision optimal energy utilization, and ensure the stability of smart grids. Most of the solutions do not exploit cross-domain monitoring capabilities, which can be implemented courtesy of the rapid proliferation of Internet-of-Things (IoT) technologies. Moreover, the recent advancements in machine learning (ML) and artificial intelligence (AI) provide a promising pathway to transcend beyond monitoring and state estimation to prediction and proactive optimizations. Lastly, the infrastructure costs for application-layer implementation can be minimized by provisioning charging station recommendation systems as a dedicated application hosted on 5G and beyond cellular infrastructure. EVOLVE’s vision is to realize innovative novel charging point recommendation systems which exploit the above-mentioned technologies. The research and innovation exchange program is based on inter-disciplinary, inter-sectoral, and inter-country mobility to realize the envisioned design goals.
47 secondments out of 150 have been carried out in the EVOLVE project so far. The work done is described below. Some of the representative are:
- UNN to SC, 1PM. This secondment was focused on developing a model that incorporates an optimization problem designed to minimize the energy cost of the Ada Byron building at UMA, which encompasses PV panels, a storage system, HVAC, and an electric vehicle charging point. To do so, a price-based control method was considered for the HVAC system. The objective function of the optimization problem encompasses three key terms: the minimization of purchased power from the upstream grid, the degradation cost of the battery, and a penalty term associated with the HVAC system.
- UOU to TAT, 5.9 PM. The secondment focused on two main objectives for public charging places: generating a three-year synthetic EV dataset for training an uncertainty-aware DL model, addressing insufficient real EV data; and designing a charging scheduling algorithm for capacity planning. ML-based models utilized the synthetic dataset, considering factors like EV growth, weather, holidays. The algorithm, including a mixed integer linear problem (MILP) and heuristic online scheduling, aimed at optimizing power demand based on predicted EV arrivals, ensuring a sustainable and efficient public charging infrastructure.
- UMA to TRU, 3.2 PM. The secondment focused on the realization of data predicting layer orchestrating distributed neural networks and implementing incremental training within the Kafka-ML framework. Success in deploying these methods enabled incremental federated training for enhanced data sharing among EVs, improving prediction and scheduling of charging times. The result of this secondment will be sent to an international congress and scientific journals.
- UNN to SC, 1PM. This secondment was focused on developing a model that incorporates an optimization problem designed to minimize the energy cost of the Ada Byron building at UMA, which encompasses PV panels, a storage system, HVAC, and an electric vehicle charging point. To do so, a price-based control method was considered for the HVAC system. The objective function of the optimization problem encompasses three key terms: the minimization of purchased power from the upstream grid, the degradation cost of the battery, and a penalty term associated with the HVAC system.
- UOU to TAT, 5.9 PM. The secondment focused on two main objectives for public charging places: generating a three-year synthetic EV dataset for training an uncertainty-aware DL model, addressing insufficient real EV data; and designing a charging scheduling algorithm for capacity planning. ML-based models utilized the synthetic dataset, considering factors like EV growth, weather, holidays. The algorithm, including a mixed integer linear problem (MILP) and heuristic online scheduling, aimed at optimizing power demand based on predicted EV arrivals, ensuring a sustainable and efficient public charging infrastructure.
- UMA to TRU, 3.2 PM. The secondment focused on the realization of data predicting layer orchestrating distributed neural networks and implementing incremental training within the Kafka-ML framework. Success in deploying these methods enabled incremental federated training for enhanced data sharing among EVs, improving prediction and scheduling of charging times. The result of this secondment will be sent to an international congress and scientific journals.
Some of the key results which advance beyond the state of the art are described below:
- One key result aimed to design efficient algorithms for multi-tier, multi-RAT networks. Active contributions at Princeton led to a technology recognition-based spectrum sensing system, identifying waveforms from diverse networks. The proposed DSS scheme does not require a coordination signaling channel between the LTE and NR networks. Instead, a technology recognition-spectrum sensing system is used to estimate traffic patterns and make spectrum allocation decisions accordingly. The goal is to enhance the seamless integration of these technologies, ensuring reliable communication and connectivity for vehicles in dynamic and varied network environments.
- To train the ML-based uncertainty-aware models, we need a certain amount of real collected EV data. Addressed the challenge of insufficient EV-related Data, particularly for capturing yearly seasonal patterns affected by holidays, events, and the growth of EVs on EV’s arrival date. To tackle this, the synthetic data is generated spanning three years (2020 to 2022) based on one year of collected ACN data. The factors influencing EV arrivals, including EV growth rate, hourly weather conditions, holidays, and events are also incorporated. This generated realistic dataset is utilized to train the uncertainty-aware prediction model.
- This result aimed at defining a general model for a digital twin ecosystem capable of representing standard EVs scenarios. To this end, a detailed investigation was carried out on the various entities that could make up this scenario, such as EVs, charging stations, and connectors, subsequently identifying specific variants and brands (e.g. BEVs, PHEVs, and EREVs for EVs). In the modeling process, digital twin types were designed, where each type abstractly represents an independent entity or variant. Likewise, the composition of digital twins was considered, allowing the decomposition of complex twins into simpler entities through parent-child relationships.
- One key result aimed to design efficient algorithms for multi-tier, multi-RAT networks. Active contributions at Princeton led to a technology recognition-based spectrum sensing system, identifying waveforms from diverse networks. The proposed DSS scheme does not require a coordination signaling channel between the LTE and NR networks. Instead, a technology recognition-spectrum sensing system is used to estimate traffic patterns and make spectrum allocation decisions accordingly. The goal is to enhance the seamless integration of these technologies, ensuring reliable communication and connectivity for vehicles in dynamic and varied network environments.
- To train the ML-based uncertainty-aware models, we need a certain amount of real collected EV data. Addressed the challenge of insufficient EV-related Data, particularly for capturing yearly seasonal patterns affected by holidays, events, and the growth of EVs on EV’s arrival date. To tackle this, the synthetic data is generated spanning three years (2020 to 2022) based on one year of collected ACN data. The factors influencing EV arrivals, including EV growth rate, hourly weather conditions, holidays, and events are also incorporated. This generated realistic dataset is utilized to train the uncertainty-aware prediction model.
- This result aimed at defining a general model for a digital twin ecosystem capable of representing standard EVs scenarios. To this end, a detailed investigation was carried out on the various entities that could make up this scenario, such as EVs, charging stations, and connectors, subsequently identifying specific variants and brands (e.g. BEVs, PHEVs, and EREVs for EVs). In the modeling process, digital twin types were designed, where each type abstractly represents an independent entity or variant. Likewise, the composition of digital twins was considered, allowing the decomposition of complex twins into simpler entities through parent-child relationships.