Non-Intrusive Load Monitoring (NILM) is particularly important for demand response. This paper proposes an innovative method based on a deep learning model to recognize the working state of electrical appliances using low frequency load data. The approach includes a data processing step, a deep learning model and a new accuracy calculation method. The data processing step consists of a multi-feature and high-dimensional method (MFHDM) and a pre-training process. The deep learning model consists of a convolutional neural network (CNN), a long-term short-term memory network (LSTM) and a random-forest (RF) algorithm. The proposed method addresses the label correlation problem and the class-imbalance problem. To test the proposed method, the Reference Energy Disaggregation Dataset (REDD) and the Pecan Street dataset (PSD) are used. A comparative analysis with several models shows that the proposed method can effectively improve electrical appliance recognition accuracy and realize NILM.
The quench of high-temperature superconducting (HTS) devices will seriously affect its safety operation. A fast and reliable quench detection method is of great significance for quench protection. In this paper, a quench detection method based on the CNN-LSTM model is proposed. By processing and analyzing the collected voltage signals in real time, the model can realize quench detection instantly and accurately. Furthermore, by adjusting the probability threshold, the sensitivity and the accuracy of the model can be balanced. Based on a series of quench experiments of HTS tapes and derived data, the CNN-LSTM model is trained and tested. The results show that the proposed method is advanced compared with the voltage threshold method.
30 March 2018 in IEEE Transactions on Applied Superconductivity
An HTS superconducting magnetic energy storage (SMES) can be utilized to improve the security and stability of the power grid with renewable energy generation. In different application scenarios, the requirements for capacity and power of the SMES varies, which needs the appropriate SMES allocation in the power grid. A cost estimation is the key component for the SMES allocation. Therefore, this paper proposed a cost estimation method of an HTS SMES, which synthetically considers the design, manufacture, operation, maintenance, and application scenarios. First, the cost estimation model of an HTS SMES was proposed based on the optimal superconducting magnet design. Then, adopting typical scenarios in the power grid, the economic capacity of SMES allocations were analyzed. Finally, the cost trend of an HTS SMES was carried out based on the development trend of assembly units.
The improvement of intelligent appliances provides the basis for the demand response (DR) of residential loads. Thermostatically controlled loads (TCLs) are one of the most important DR resources and are characterized by a large load and a high degree of control. Due to its distribution characteristic, the aggregation of TCLs and their control are key issues in implementing the load control for the DR. In this study, we focus on air conditioning loads as an example of TCLs and propose a simple and transferable aggregate model by establishing a virtual house model, which accurately captures the aggregate flexibility. The deviation of the aggregate model is analyzed for the model evaluation. An air conditioning DR control scheme is proposed based on the aggregate model; it has the advantage of simple implementation and convenient control for the individual units. Simulations are performed in Gridlab-D to evaluate the accuracy and effectiveness of the proposed model and control method.
Deep learning plays an important role in short-term load forecasting. By combining deep learning model with expert knowledge, the accuracy and interpretation of load forecasting model can be further improved. Under the guidance of load forecasting theory, the effect of holidays on daily load was taken into account. This paper proposes a new convolutional LSTM (ConvLSTM) deep learning model with taking advantage of expert knowledge. Simulation results show that the proposed ConvLSTM network captures spatiotemporal correlations better and consistently outperforms LSTM for load forecasting, especially in holidays load forecasting.