Application Research Of Machine Learning In Power Load

The collection and storage of large-scale load data in smart grid provide new approaches for the efficient,economical,and safe operation of power systems.Systematic analytics of these large-scale load data and proper management can ensure uninterrupted power supply with minimum cost.Machine learning(ML)has been in use for load data because of its ability to discover complex relationships.With the goal to ensure uninterrupted power supply with minimum cost,the application research of ML techniques on power load can be categorized into three major areas: analysis,forecast,and management.All the three major application areas of ML in power load,should be considered equally.This thesis is equipped with the aim of solving existing challenges in three application areas.To find the research gap and identify key challenges in three application areas,review of the current research condition played vital role.In this thesis,three challenges have been identified for the three application areas.(1)Among various issues of load analysis,identification of load affecting factors is a burning issue.The rapid growth of electric vehicles(EVs)is likely to danger the current power system.Forecasting the demand for charging stations is one of the critical issues while mitigating challenges caused by the increased penetration of EVs.Uncovering load-affecting features of the charging station can be beneficial for improving forecasting accuracy.It is not possible to consider all the load affecting factors because diverse factors involved in load changing.Even though we know the factors,we can not include all of these factors because of the lack of information on many of these causes.As a result,modeling an accurate forecasting model for shortterm load is still challenging.After getting the accurate forecast,the next issue for the power operator is to balance the peak and valley of the power demand for optimal operation,which falls into the field of power management.From the management perspective,power system faces a lot of challenges.Among them,challenges related to plug-in electric vehicles(PEVs)inclusion are threatening the stability of the existing power system.The encouragement and subsidy scheme taken by governments are fueling the use of PEVs.In addition to the charging station,hassle-free in-house charging is getting popular.Due to high power consumption,in-house PEV charging has a significant impact on the distribution network,and the utility companies are facing challenges in balancing the power demand.(2)Existing studies on electricity demand forecast of charging station is mostly based on load profiling.It is difficult for public EV charging stations to get features for load profiling.This thesis examines the power demand of two workplace charging stations to address the above-mentioned issue.Eight different types of load-affecting features are discussed in this study without compromising user privacy.Later the features are used to design forecasting model.Finally,a state-of-the-art interpretable machine learning technique has been used to identify top contributing features.To mitigate the effect of unknown and unavailable load affecting factors,error trend is used and proposed adaptive second learning of error trend(A-SLET).Furthermore,the training set is classified based on balance point temperature and then parallelly trained and tested adaptive forecaster for hot days and adaptive forecaster for cold days with proper data.Combining A-SLET with parallel forecasting and training set classification,Adaptive and Parallel forecasting strategy based on Second Learning of Error Trend(AP-SLET)is proposed.The work studied two distinct load patterns,one in the USA and the other in Australia.The problem of increased power demand due to inclusion of PEV could be solved by increasing distribution network capacity,designing an effective demand response program aimed at PEV charging,implementing vehicle to grid(V2G),or vehicle to house(V2H),and so on.One of the fundamental parameters for the mentioned solutions is to identify households with PEVs.Multi-level charging,power consumption similar to other home appliances,and absence of submeter for charging outlets make identification difficult.This thesis proposes a new feature extraction technique called knowledge-based systematic feature extraction to identify households with PEVs.(3)The study on charging station load demand finds a few load changing factors that are crucial to improving load forecasting accuracy.We found that the workplace EV charging station exhibits opposite characteristics to the public EV charging station for some factors.Moreover,using state-of-the-art interpretable machine learning techniques,we found number of session,session duration,and idle occupancy are the three top common demand affecting factors.As the study is conducted on a publicly available dataset and analyzes the root cause of demand change,it can be used as baseline for future research.Experimental results using proposed AP-SLET find that MAPE is 1.87 %-4.04 % for ME-Maine of New England and 2.81 %-4.41 % for New South Wales across a yearly forecasting horizon.Compared to the state-of-art forecasting methods,MAPE of the AP-SLET is reduced by 17.03%-33.33%,RMSE and MAE are reduced by 34.05% and 35.38%,respectively.The experimental results demonstrate the proposed strategy can transform unknown and unavailable load affecting factors into known forecasting features and then adapt it to improve forecasting performance.The proposed strategy is also forecaster independent and equally applicable to almost all load scenarios regardless of geographical and seasonal differences.The extracted features for identifying households with PEVs are easily interpretable and validated using two datasets from different regions.Keeping real scenarios in mind,the study examines scenario-based results and finds that the accuracy using extracted features ranges from 80.20% to 100%,depending on classifier,number of vehicles,and level of charging.Moreover,results show improved performance compared to existing methods for identifying households with PEVs and other state-of-the-art feature extraction techniques.Overall,the techniques proposed in this thesis show effectiveness in experiments and can be used in power systems.