Home energy monitoring by appliance-level information can provide consumers awareness on energy saving. The system can be implemented through a smart meter which requires an efficient data analysis algorithm for providing an accurate energy consumption profile, the purpose for proper home energy management. This paper proposes a set of data analysis procedures for extracting appliances power state from its power consumption data. The approach is based on multitarget classification, a new data learning framework for nonintrusive load monitoring. The procedures include 1) partitioning the appliance power data into an effective number of power states using K-means clustering. 2) determining the optimal number of power states using the Area Under the ROC Curve index. The design objective is to obtain the optimal predictive performance for identification of the appliance power state which could result in a proper power and energy prediction. Applying the multitarget classification algorithm of RAndom k-labELsets by disjoint subsets with the decision tree, the identification of appliance power state achieved F-score and accuracy values greater than 89% for high-power loads such as A/C and water heater. The normalized error values of power prediction outperformed the use of Factorial Hidden Markov Model and binary state modeling system.
This paper proposes an experimental design process for the application of energy disaggregation using multi-target regression, a new data learning approach in this application area. The approach shows to be a suitable model for dealing with energy disaggregation problems in which the task is to predict multiple appliances usage from the aggregate data. The experiments were conducted by analyzing AMPds2 and ECO public data sets for verifying the effectiveness of the approach. The data were analyzed through the machine learning process to select the optimal set of electrical features, learning algorithm, and model parameter so that the system resulting from the process could deliver the optimal performance for loads inference. Results of the data learning showed that the electrical features set of current ( $I$ ), real power ( $P$ ), reactive power ( $Q$ ), and power factor ( PF ) for the aggregate data and Random Forest as the base regressor for multi-target regression model could provide the best disaggregation performance. The overall predictive performance of disaggregation accuracy and F-score outperformed the benchmarking Super State Hidden Markov Model (SSHMM) and Denoising Autoencoder (DAE) network approaches.
Insightful information on energy use encourages home residents to conduct home energy conservation. This paper proposes an experimental design for an energy disaggregation system based on the low-computational-cost approaches of multi-target classification and multi-target regression, which are under the multi-target learning framework. The experiments are set up to determine the optimal learning algorithm and model parameters. In addition, the designated system can provide inference of the appliance power state and the estimated power consumption from both approaches. The kernel density estimation technique is utilized to formulate the appliance power state as a finite-state machine for the multi-target classification approach. Multi-target regression can directly provide the estimation of appliance power demand from the aggregate data, and this work unifies the system’s design together with multi-target classification. The predictive performances obtained through the F-score (micro-averaged) and power estimation accuracy index for the power state inference and the estimated power demand, respectively, are shown to outperform a deep-learning-based denoising autoencoder network under the same data settings from both approaches. The results lead to a recommendation to apply the approach in home energy monitoring, which is mainly based on the characteristics of appliance power and the information that the residents wish to perceive.
This paper describes a dataset for evaluating the effectiveness of energy disaggregation techniques, specifically Non-Intrusive Load Monitoring (NILM) algorithms. The dataset comprises real-world energy consumption data from a residential household in Bangkok, Thailand. It includes one aggregate circuit and ten active sub-circuits, including lights, plug-outlets, a water pump, a water heater, and an air conditioner, sampled at a one-minute interval. The dataset provides voltage, current, active power, reactive power, and power factor measurement data, facilitating experimentation with multiple energy disaggregation scenarios. The inclusion of the absolute and signed reactive power (Q) data allows greater flexibility in feature selection for model training. This dataset has been validated and utilized in previous research, demonstrating its robustness and high accuracy in appliance classification and power state detection. The data can also be applied to diagnose appliance faults and assess energy efficiency.
Accurate identification of appliance operations is crucial for effective energy management in smart energy monitoring or energy disaggregation. This paper proposes a multi-label classification approach that utilizes adaptive feature integration of steady-state features (current (I), active power (P), reactive power (Q), power factor (PF)) and steady-state change features (ΔI, ΔP, and ΔQ) obtained from low-frequency meter data. Recursive Feature Elimination with Cross-Validation (RFECV) is employed as the feature selection technique to determine the optimal subset of features for enhancing individual appliance identification. Experiments are conducted on two datasets, AMPds and ECO, consisting of high-power appliances that consume a significant amount of power when turned on (e.g., Wall Oven, Clothes Dryer, Kettle) and low-power appliances that consume less power (Fridge, Freezer). Results demonstrate significant performance improvements on the AMPds dataset by the additional selective features, increasing the F-score value for the Wall Oven from 0.56 to 0.70, outperforming Forward Sequential Feature Selection (FSFS) as a feature selection method comparator and a benchmarking Denoising Autoencoder (DAE) neural network model. In contrast, the ECO dataset yields marginal performance improvement for most appliances due to subtler steady-state change signatures. Additional evaluations on sensitivity analysis indicate that the approach can provide superior performance compared to using the baseline features under the data sampling frequencies of 5 and 10 minutes, noisy levels of 1% and 5%, and appliance concurrency up to 3 appliances. The proposed approach of deploying basic features and adapting to variations in measurement setups is suitable for use in resource-constrained environments.