Power Management by LSTM Network for Nanogrids

Abstract

Nanogrids can be considered smart grids that are implemented for small-scale buildings, houses, and apartments. A typical power management framework for nanogrids determines the scheduling of operations of electric appliances for each time interval with objectives related to total power consumption and total delay due to scheduling. Such a framework of power management has limitations in accommodating future operating conditions of nanogrids. Taking future outdoor temperature as a future operating condition, a proactive power management for nanogrids is presented in this paper. The goal of proactive power management for nanogrids is to achieve the proper level of indoor temperature in a cost-efficient way, sooner rather than later, by taking into account future outdoor temperature. To achieve this goal, a long short-term memory (LSTM) network is used as the controller. Simulations have been performed to verify the performance of the proposed power management. The results of the simulations demonstrate that living comfort measured in terms of room temperature is enhanced while the overall electricity cost is reduced, mainly due to the ability of the LSTM network to predict the trend of outdoor temperature.

Management of Distributed Renewable Energy Resources with the Help of a Wireless Sensor Network

“…The ML and artificial intelligence technologies have been applied in forecasting as computer processing speed becomes faster. The ML methodologies, as compared to physical forecasting approaches, typically produce better outcomes [43,58], but on the other hand, necessitate a significant amount of data during the training process and a model with a vast range of training data is easy to overfit. Many researchers in recent years have applied a variety of means and tactics to achieve high forecasting accuracy.…”

Design of a Smart Nanogrid for Increasing Energy Efficiency of Buildings

“…Thus, by utilising a PV system, a BESS and lifestyle behavioural changes that result in load shifting, the grid usage has been reduced from 14.5 MWh/year to 1.17 MWh/year. On the other hand, instead of requesting the user's input for load flexibility, the thermal elements can be utilised to perform DSM without the occupant's interaction [30]. Thus, in order to enhance the living comfort of the building, the indoor temperature has been used as the input parameter for controlling the available heating and cooling elements.…”

“…The results of the algorithm show that the equilibrium point is reached as the nanogrid population increases, demonstrating the effectiveness of interconnected nanogrids. In addition, a decentralised droop control method has been used in [30], which uses the SoC of the battery of neighbouring DC nanogrids to facilitate power exchange between them. This method does not require communication between the nanogrids, as the power exchange is initiated based on a scheduled voltage threshold of the battery.…”

Short-Term Predictive Power Management of PV-Powered Nanogrids

“…The training parameters of the LSTM network, shown in TABLE 3, are as follows. The total number of training epochs is 1000 and batch size is 200 and the ADAM [56] optimization algorithm is used with learning rate 0.005, gradient moving average 0.9, dropout rate 0.2, and gradient threshold 1 [36], [57]. Inadequate learning rates might cause an undesirable local minimum and overfitting.…”

“…To this end, artificial neural networks such as long shortterm memory (LSTM) [34] network can be adopted. The LSTM network has been used for various applications [35], [36]. According to the forecasting horizon, the forecasting model can be categorized into three categories: short-term forecasts upto one week, medium-term forecasts from one week to one year, and long-term forecasts over one year.…”