The emergence of microgeneration systems steadily increases, and it raises concerns regarding their impact on the power grid. It is, therefore, crucial to efficiently integrate them into future smart grid architectures, as there is not any standard way to monitor production units. Moreover, current data collection systems are simple and do not consider their impact on local area networks. This paper presents a set of proposed mechanisms that reduces the monitoring traffic, while offering management flexibility on large-scale systems. This study is illustrated with measurements performed on a small grid, and it shows that, for monitoring a photovoltaic production, both 1-min and 1-s intervals provide the same production estimation, while significantly decreasing the associated traffic. It can be reduced even more by aggregating several measurements during a given period before sending them and by using specific mechanisms to ensure reliability. This experiment also helps authors identify best practices for monitoring different equipment based on their behaviors.
With the increasing concerns about privacy and data regulations, federated learning (FL) has been emerging as a solution to train machine learning models collaboratively with non-exchangeable data from multiple clients. As a result of data locality, data is usually not identically or independently (non-IID) distributed across clients, and the non-IID property has long been the key challenge in FL. Furthermore, in real-world cross-silo scenarios, it is ubiquitous that clients are organizations owning private data from multiple domains internally, which exacerbates the non-IID issue. For example, in healthcare applications, each client (hospital) gathers data from patients with heterogeneous demographics. While previous works have made efforts to address the non-IID challenge across clients by assuming various relations among client-level data distributions and enabling personalized models at the client level, they ignore the internal data heterogeneity within each client or require explicit data domain indicators, which are hardly accessible in real-world data. Here, we propose Sample-Level Prototypical Federated Learning (SL-PFL) to bridge the gap. SL-PFL incorporates prototypical learning under the FL framework and provides a fine-grained personalized model for each data sample instead of learning one uniform model for all samples of each client. Meanwhile, it can be trained using data without ground-truth domain indicators. Experimental results demonstrate that our proposed method with sample-level personalized models outperforms existing FL methods with a global model or client-level personalized models on various real-world regression and classification tasks from weather, computer vision, and healthcare applications.
The number of connected devices - also known as Internet of Things (IoT) - is exponentially increasing. Such sensors and devices also appear in transportation systems giving some intelligence to roads, equipment and vehicles. Nowadays, it is possible to communicate with the environment in order to have better everyday services. Furthermore, the number of registered - public or private - Electric Vehicle (EVs) is continuously increasing. These vehicles, equipped with large battery, need to be charged and so, have a significant impact on power grids. However, these EVs can also be seen as energy sources. It is therefore important to be able to plan both the charge and discharge of EVs. Including these vehicles into Vehicle-to-Grid technology is a way to efficiently manage such pools of batteries. But, as a consequence, grid requires to have almost real-time data on these vehicles and especially their battery status. This paper studies an optimized data aggregation method for a fleet of electric buses. Each bus provides different type of information with different priority level. The efficiency of the studied method was evaluated with a simulation platform developed with ns-3. Simulation results - based on real route and bus stop positions - show that an optimal buffer size has been found to both satisfy transmission delays and optimize communications.
The fifth edition of Sussex-Huawei Locomotion-Transportation (SHL) recognition challenge aims at determining the mode of locomotion and transportation of users using sensor-equipped devices. This recognition task relies on measurements captured from Inertial Measurement Unit and radio sensors placed on the user’s hand, bag, hips and torso. The provided dataset presents several challenges, including its size, asynchronicity between the data types, lack of time continuity, and imbalanced distribution of the locomotion and transportation modes. To address these issues and enhance the recognition performance, our team, KDDI Research, performed data pre-processing and hand-crafted additional features. We explored different classifiers based on both Machine Learning or Deep Learning methods. Finally, the XGBoost Classifier model achieved the highest accuracy and f1-score across different validation datasets (bag, hands, hips, and torso). This model, used for our final submission on the testing dataset, achieved an average accuracy of 0.75 on these datasets.