Table 4.
Qualitative comparison of smart grid related works.
Use Case | Ref | Contribution | AI Role (At the Edge) |
AI Algorithm | Dataset | AI Placement | Employed Technology |
Platform | Metrics | Benefits AI-Edge |
Drawbacks | |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Smart grid | LDF | [33] | Short-term energy consumption forecasting | Prediction | LSTM | Pecan Street Inc’s Dataport site | Edge, cloud | Federated learning | Python, TensorFlow Federated 0.4.0 Tensorflow 1.13.1 backend | RMSE, MAPE | High accuracy | Heterogeneous data unsolved |
[34] | Short-term energy consumption forecasting | Prediction, classification | LSTM, K-means | Energy company UK Power Networks | Edge device, cloud | Federated learning | Python, TensorFlow | RMSE, training time | High accuracy, heterogeneous data solved | Privacy still low | ||
[35] | Day-ahead prediction of building energy demands | Prediction, Feature selection | Ant-bee, cuckoo, elephant, flower, genetic harmony, PSO, rhino, wolf, DT, HT | Ornl-research-house-3 | Edge server (Raspberry Pi) | Low-cost model | Keras, Python | Accuracy, time, speed, MAE | High accuracy, low training time | Low interpretability | ||
[36] | Short-term electricity demand | Prediction, classification | XGBoost, K-means | Tianchi under license | Edge server (PC) | Low-cost model | Not mentioned | Training time, accuracy, cross-entropy loss | High accuracy | Data distribution unsolved | ||
[37] | Short-term electricity demand | Outlier detection, Feature selection, prediction | NB, wrapper FS, Filter FS | EUNITE dataset | Fog nodes | Matlab | Accuracy, error, precision, sensitivity/recall | High accuracy, reliability, resilience, stability | High complexity of model | |||
[38] | Online short-term energy prediction | data preprocessing, prediction | DNN | Real-world dataset | Edge server, edge devices, cloud | Collaborative learning | Not mentioned | Flexibility, accuracy | Flexibility, high accuracy, dynamic data, IoT addressed, real-time prediction | Less scalability | ||
[39] | Load forecasting for optimal energy management | Prediction | CNN | IHEPC dataset | Edge devices | / | TensorFlow, Keras | MAPE, RMSE | Low complexity | Heterogeneous data, uncertainties, privacy is not addressed | ||
[40] | Online short-term residential load forecasting | Prediction | STN | Ohta-AMPds datasets | Edge device | Low-cost model-reservoir computing | Not mentioned | RMSE, MAE | Low complexity, high accuracy | Heterogeneity not addressed | ||
D.S.M | [41] | Demand-side management | Resource management | RL | Real-world dataset | Edge server (Raspberry Pi) | – | Real implementation | Not mentioned | / | Less scalability | |
[42] | Demand-side management | Classification | LDA | REFIT project | Edge server | Low-cost model | Not mentioned | MAPE, RMSE | ||||
[43] | Managing prosumers over wireless networks | Data preprocessing, prediction | LSTM | Pecan Street Inc.’s Dataport site | Edge server | Federated learning | TensorFlow | RMSE, data transmitted | Heterogeneous data addressed, high accuracy low-communication cost | Single-point failure not addressed | ||
LAD | [44] | Detection of anomalous power consumption at household | prediction | GBR, RFR, LR, SVR | IHEPC dataset | Edge server, fog | / | Not mentioned | MAPE, RMSE | Load reduction | Communication cost still high | |
[45] | Anomaly detection in smart-meter data | resource allocation, classification | SDA, GA, kNN | IHEPC dataset | Edge server | / | Not mentioned | Accuracy, execution time, energy consumption | – | – | ||
[46] | Electric energy fraud detection | Dimensionality reduction, prediction | DTR, LR | D1C database | Edge server Raspberry Pi model | – | Not mentioned | MAPE | – | – | ||
[47] | Anomaly detection consumption smart grid | Classification | DNN, HDBSC K-means, KNN | Midwest region | Edge server, Raspberry Pi | / | Not mentioned | Testing time, frequency, model size | Low complexity, high accuracy | – | ||
[48] | Energy theft detection | Feature-extraction classification | VAE-GAN, K-means | GEF Com 2012 public dataset | Edge server | / | Not mentioned | ROC curve, running efficiency | Adaptive model, high accuracy | - | ||
[49] | Energy theft detection | Classification | (SGCC) dataset | Edge devices | Federated learning | Flower | RMSE, log loss accuracy, precision F-measure | Privacy | Low accuracy compared with the centralized model |