Federated learning (FL) has been proposed as a novel paradigm to enable distributed learning on theedge with privacy protection. However, existing federated learning approaches mainly focus on training deep classific...
详细信息
ISBN:
(数字)9781665480185
ISBN:
(纸本)9781665480185
Federated learning (FL) has been proposed as a novel paradigm to enable distributed learning on theedge with privacy protection. However, existing federated learning approaches mainly focus on training deep classification and clustering models, and no enough attention has been paid to solve the federated reinforcement learning task on theedge, a challenging task where multiple learning agents observe local state and take local actions to train a global learning model without revealing their local dataset. In this paper, we propose a generalised federated reinforcement learning framework called FedMC that integrates reinforcement learning models trained by multipleedge devices into a general model based on a metalearning approach. In the proposed framework, each participant adopts a meta-value network (MVN) and task-actor encoder network (TAeN) locally to perform meta-learning training based on local task samples, and periodically uploads the weights of local MVN and TAeN to the server, which aggregate them to a global model with rapid adaptability and cross-task applicability. extensiveexperiments based on a number of reinforcement learning tasks show that FedMC outperforms various federated learning baseline algorithms, and it is competitive with the methods that centrally train reinforcement learning task with global dataset. Index Terms-
This paper takes the two-machine and two-zone DFIG system as the research object, and studies two virtual synchronous control strategies of the "voltage and current double loop control strategy"with current ...
详细信息
The integration of energy flow, information flow, and service flow is essential for the distribution substation area intelligence and is important for the development of the new power system. The common information mo...
详细信息
Parallel computing of real-time data is an effective solution to the problem of timely analysis with the rapid increase of real-time data in today's power grid. Currently, most data parallel processing methods do ...
详细信息
Deploying edge servers on Unmanned Aerial Vehicles (UAVs) has become a promising strategy to handle the spatiotemporally varying user demands on ground. However, a critical limitation of the UAVs is the limited flight...
详细信息
In theexpanding field of quantum computing (QC), efficient and seamless integration of QC and high performancecomputing (HPC) elements (e.g., quantum hardware, classical hardware, and software infrastructure on both...
详细信息
This study compares discrete pulse width modulation (PWM) and improved hysteresis PWM controllers for voltage sag mitigation in grid-connected hybrid PV-wind power systems with dynamic voltage restorers (DVRs) based o...
详细信息
This study offers experimental testing of commercial and laboratory inverters, utilized in a laboratory prototype of an urban microgrid. Operation of grid inverters supplied by PV arrays in urban environments, encount...
详细信息
To progress towards sustainable agriculture, crop production processes have to be optimized. Avoiding the use of herbicides is a major issue and a way to go on the agro-ecological transition;hence the interest in robu...
详细信息
Smart grid is distributed through gridcomputing programs that focus on theefficient operation of power systems in applications to be used in conjunction with. Involvement in SGS techniques involves threats such as p...
详细信息
暂无评论