In a conventional power grid, energy theft is difficult to detect due to limited communication and data transition. The smart meter along with big data mining technology leads to significant technological innovation in the field of energy theft detection. This paper proposes a convolutional long and shortterm memory (ConvLSTM) based energy theft detection (ETD) model to identify electricity theft users. In this work, electricity consumption data is reshaped quarterly into a 2-dimensional matrix and used as the sequential input to the ConvLSTM. The convolutional neural network (CNN) embedded into the long short-term memory (LSTM) can better learn the features of the data on different quarters, months, weeks, and days. Besides, the proposed model incorporates batch normalization. This allows the proposed ETD model to support raw format electricity consumption data input, reducing training time and increasing the efficiency of model deployment. The final experimental results show that the proposed ConvLSTM model exhibits good robustness. It outperforms the multilayer perceptron (MLP) and CNN-LSTM in terms of performance metrics and model generalization capability.

Accurate short-term load forecasting is essential to the modern power system and smart grids; the utility can better implement demand-side management and operate the power system stable with a reliable forecasting system. The load demand contains a variety of different load components, and different loads operate with different frequencies. Conventional load forecasting models (linear regression (LR), Auto-Regressive Integrated Moving Average (ARIMA), deep neural network, etc.) ignore frequency domain and can only use time-domain load demand as inputs. To make full use of both time domain and frequency domain features of the load demand, a hybrid component decomposition and deep neural network load forecasting model is proposed in this paper. The proposed model first filters noises via wavelet-based denoising technique, then decomposes the original load demand into several sublayers to show the frequency features while the time domain information is preserved as well. Then bidirectional LSTM model is trained for each sub-layer independently. To better tunning the hyperparameters, a Bayesian hyperparameter optimization algorithm is adopted in this paper. Three case studies are designed to evaluate the performance of the proposed model. From the results, it is found that the proposed model improves RMSE by 66.59% and 84.06%, comparing to other load forecasting models.

The integration of small-scale PV systems (such as roof-top PVs) decreases the visibility of the power system since the real demand load is masked. Most of the rooftop systems are behind-the-meter and cannot be measured by the household smart meter. To overcome the challenges mentioned above, this paper proposes an online solar energy disaggregation system to decouple the solar energy generated by the roof-top PV systems and ground truth demand load from the net measurements. A 1D CNN bidirectional long short-term memory (CNN-BiLSTM) deep learning method is used as the core algorithm of the proposed system. The system takes a wide range of online information (AMI data, meteorological data, satellite-driven irradiance, and temporal information) as inputs to evaluate the PV generation, and the system also enables online and offline modes. The effectiveness of the proposed algorithm is evaluated by comparing it to baselines. The results show that the proposed method reaches good performance under different penetration rates and different feeder levels. Finally, a transfer learning process is introduced to verify the proposed system has good robustness and can be applied to anywhere else easily.

The purpose of feeder-level energy disaggregation is to decouple the net load measured at the feeder-head into various components. This technology is vital for power system utilities since increased visibility of controllable loads enables the realization of demand-side management strategies. However, energy disaggregation at the feeder level is difficult to realize since the high-penetration of embedded generation masks the actual demand and different loads are highly aggregated. In this paper, the solar energy at the grid supply point is separated from the net load at first via either an unsupervised upscaling method or the supervised gradient boosting regression tree (GBRT) method. To deal with the uncertainty of the load components, the probabilistic energy disaggregation models based on multi-quantile recurrent neural network model (multi-quantile long short-term memory (MQ-LSTM) model and multi-quantile gated recurrent unit (MQ-GRU) model) are proposed to disaggregate the demand load into controlled loads (TCLs), non-thermostatically controlled loads (non-TCLs), and non-controllable loads. A variety of relevant information, including feeder measurements, meteorological measurements, calendar information, is adopted as the input features of the model. Instead of providing point prediction, the probabilistic model estimates the conditional quantiles and provides prediction intervals. A comprehensive case study is implemented to compare the proposed model with other state-of-the-art models (multi-quantile convolutional neural network (MQ-CNN), quantile gradient boosting regression tree (Q-GBRT), Quantile Light gradient boosting machine (Q-LGB)) from training time, reliability, sharpness, and overall performance aspects. The result shows that the MQ-LSTM can estimate reliable and sharp Prediction Intervals for target load components. And it shows the best performance among all algorithms with the shortest training time. Finally, a transfer learning algorithm is proposed to overcome the difficulty to obtain enough training data, and the model is pre-trained via synthetic data generated from a public database and then tested on the local dataset. The result confirms that the proposed energy disaggregation model is transferable and can be applied to other feeders easily.

The advanced metering infrastructure (AMI) is a compulsory component of the future smart grid; it not only provides near real-time two-way communication between the consumers and the utility but also gives an opportunity to third parties to provide relevant value-added services to the consumers to improve the user satisfaction. However, existing services require the consumers share their private energy data with other parties, which has potential privacy risks. To better balance the excellent quality of the services and privacy guarantee, a novel differential private federated learning-based third-party service platform is proposed. Instead of sending the original energy data to the cloud server, the central server in the proposed scheme only collects the model parameters, which are trained locally inside the consumers’ houses. Then the collected parameters are aggregated differential privately to eliminate the identity of individuals, and the aggregated parameters are used to update the central model and improve the model performance. Furthermore, a novel attention-based bidirectional long short-term memory neural network model is adopted to make predictions. In the case study, a residential short term load forecasting task is implemented to evaluate the performance of the proposed model; from the simulation results, the conclusion is made that the proposed model can achieve similar accuracy as the typical centralized model and better control the privacy loss flexibly at the same time.