Posted on Leave a comment

GRL-based residential electricity behavior

Chen, X., Yu, T., Pan, Z., Wang, Z., & Yang, S. (2023). Graph representation learning-based residential electricity behavior identification and energy management. Protection and Control of Modern Power Systems8(1), 28., doi.org/10.1186/s41601-023-00305-x

Home energy management system – HEMS is an end-user energy conserving and emission-reducing method. but user behavioral identification and energy management strategy are issues of efficient HEMS and current HEMS assume user behaviour or ignore the relation between the user and appliance it causes poor management approach and improper behavioural explanation to counter these and in order to enhance HEMS decision making, this research suggests combining graph reinforcement learning (GRL) with non-intrusive load monitoring (NILM). The user behaviour and energy consumption are identified through NILM by a correlation graph and a multi-label classification approach is employed to monitor loads.

Limitations in other methods:

Research that has already been done either uses additional intrusive devices to collect user behaviour data or assumes that usage behaviour is known beforehand. These methods are not practical consideri8ng dynamic change of behaviour Conventional NILM techniques are limited to load disaggregation; they are unable to identify behaviour. The current NILM approaches have increased equipment need and poor disaggregation accuracy. Thus, the author notes that creating a practical and precise online energy behaviour detection technique for HEMS input still a difficult challenge. Certain studies make the assumption that behaviour is well-known, yet they don’t say where they got their behavioural data. However, these studies do not adequately consider the dynamic uncertainty of user According to several research, the decision-making process for improving each appliance is independent of the others. In several studies, behaviour recognition includes label correlations; still this integration is dependent on the time series signal for correlation capture of applications.

To compensate the above shortcomings the author proposes an intelligent HEMS technique uses NILM-assisted graph reinforcement learning (GRL) for behavioural detection and strategy to minimize electric cost.

Methodology:

   HEMS framework  as discussed in the study by Chen et al. (2023).

The suggested method’s practical implementation is depicted in the centre of Figure above. The ML-SGN model performs load disaggregation and behaviour identification using aggregated data from an outdoor electricity meter. In order to help customers control their home energy usage, A set of instructions is produced using the HEMS method depending on the objective function, load states, and surrounding circumstances. To be more precise, the load state relates to the online status of loads, the relationship state refers to the correlation data in the graph, and the environment state refers to external data like the cost of electricity and the temperature outside.

This method’s behavioural correlation matrix, can be computed as:

                                                                                                            pi,j = p(ai|aj) = Ni/N

The likelihood that appliance ai will function after appliance aj is represented by pi, j. The total number of times that appliance aj is on is represented by Nj, and the number of times that appliance ai operates following aj is shown by Ni, j.

Behaviour identification allows for the determination of the probability of appliance utilization during each period and the behaviour correlation. In contrast to the conventional NILM, behaviour identification requires not just determining the gadget, but moreover to extract the usage behaviour of the equipment.

Formulating the problem for HEMS:

Residential loads are separated as photovoltaic distribution (PV), loads transferable (TL), loads that can’t be controlled (UL), inaccessible loads (IL), and thermostatically regulated loads (TCL). where several appliances may be included by TCL.

The water heater equivalent thermodynamic model is expressed as follows:

Certainly, here’s a simpler version of the formula:

TWH_{n+1} = (TWH’_{n+1} * (V – V_{n,demand}) + TWH_{n,inject} * V_{n,demand}) / V

The price of energy use, comfort of users, and the consequence of breaking the restrictions, which can be expressed as

Here’s a simplified version:

rn = Ro + Cn – αEn Sn ∈ Sc

where En, Cn represent the energy consumption-related cost, comfort, and a penalty respectively.

Performance evaluation:

Six houses’ consumption of energy is provided by the REDD dataset, while twenty families’ electric power measures are provided by the REFIT dataset.

The remaining REDD and REFIT data are utilized to build a previous behaviour connection using the ML-SGN model.

Graphical correlation as discussed in the study by Chen et al. (2023).

The suggested ML-SGN model is compared with the classical multi-label classification model, multi-label k-nearest neighbour algorithm (MLKNN), random k-label sets algorithm (RAKEL), and single label model (SGN) in order to demonstrate that it not only performs better than the conventional multi-label classification methods but also shows exhibits a stronger understanding ability than original single-label models. Its effective performance can be seen by comparison with the load division with attention model (LDWA), which is baseon SGN and more sophisticated due to the attention approach.

The results of behaviour identification as discussed in the study by Chen et al. (2023).

Conclusion:

  • The suggested approach builds and modifies a graph that records customers’ power use patterns, then uses an enhanced multi-label NILM technique to pinpoint such patterns.
  • Simulations are used to assess the suggested approach, showing its improved performance in HEMS and behaviour detection.
The suggested approach offers these two key benefits:
  • In the studies, it attains a high average recognition accuracy of 93.2%, proving its efficacy in behaviour identification.
  • It keeps consumer comfort and satisfaction levels high while lowering average power costs for users by 18.3%
  • The technology offers a better balance between energy cost and user comfort than earlier approaches.

Sakthivel R

I am a First-year M.Sc., AIML student enrolled at SASTRA University

I possess a high level of proficiency in a variety of programming languages and frameworks including Python. I have experience with cloud and database technologies, including SQL, Excel, Pandas, Scikit, TensorFlow, Git, and Power BI

Leave a Reply

Your email address will not be published. Required fields are marked *