Real-time electricity price forecasting affects both the interests of power companies and the stability of power systems. Although deep learning models have achieved rich results in forecasting, due to the variable temporal characteristics and numerous influencing factors of real-time electricity prices, it is difficult for general deep learning models to extract electricity price features with obvious regularity, which affects forecasting accuracy. To solve this problem, this paper proposes an attention mechanism multi-size depthwise convolutional long short-term memory neural network (AM-MDC-LSTM) for predicting real-time electricity prices. The model improves prediction capability in the following aspects. 1) Using an attention mechanism to adaptively assign weights to electricity price time series and electricity price exogenous variables (production, consumption, electricity prices in neighboring regions) to improve electricity price feature extraction efficiency. 2) Using convolution kernels of different sizes to convolve individual electricity price exogenous variables one by one to extract local burst and global periodic electricity price features with obvious regularity. This is then combined with long short-term memory networks to extract temporal features reflected in electricity prices. Experimental results conducted in the Nordic and PJM electricity markets demonstrate that the proposed model outperforms other models discussed in the paper, exhibiting higher prediction accuracy.

Taking into account the orientation and distance characteristics of wind turbine stations in wind farms can improve the accuracy of wind power prediction. This paper proposed a deep learning spatio-temporal prediction method named orthogonal wind direction transformation spatio-temporal gradient Regression Activation Mapping (OWT-STGrad-RAM) for wind speed prediction. The model encodes the wind farm using an image, and each wind turbine is encoded as a point in the image. The spatio-temporal data related to wind turbines, such as wind speed, temperature, and air pressure, are integrated into fusion features through spatio-temporal fusion convolutional networks model for pre training to obtain a feature dataset. OWT is used to eliminate the effects of different prevailing winds, and STGrad-RAM is used to characterize the orientation and distance between wind turbine nodes and make the spatial features interpretable. The feature dataset is used for wind speed prediction. The experimental results show that the proposed method has achieved a significant improvement in wind speed prediction accuracy compared to the comparative models.

To maintain power system stability, accurate wind speed prediction is essential. Taking into account the temporal and spatial characteristics of wind speed in an integrated manner can improve the accuracy of wind speed prediction. Considering complex nonlinear spatial factors such as wake effects in wind farms, a deep residual network is valuable in predicting wind speed with a high degree of accuracy. Wind speed data are typically a time series that requires feature extraction and attribute modeling, while maintaining signal integrity. In order to measure the importance of different temporal attributes and effectively aggregate temporal and spatial features, we used a parameter fusion matrix. We introduce a deep spatial-temporal residual network (DST-ResNet) for wind speed prediction that extracts the spatial-temporal characteristics, which can forecast the future wind speed of a multi-site wind farm in a particular region. In this model, wind speed data's nearby property and periodic property are separately modeled using a residual network. The outputs of the two temporal components are dynamically aggregated using a parameter fusion matrix and then fused with additional meteorological features to achieve wind speed prediction. Based on wind data from the National Renewable Energy Laboratory, our experiments show that the proposed DST-ResNet improves prediction accuracy by 8.90%.

Considering the massive influx of new energy into the power system, accurate wind speed prediction is of great importance to its stability. Due to the influence of sensor accuracy and harsh natural environments, there is inevitable noise interference in original wind speed data, which adversely affects the accuracy of wind speed prediction. There are some problems associated with traditional signal processing methods when dealing with noise such as signal loss. We propose the use of a deep residual shrinkage unit based on soft activation (SDRSU) in order to reduce noise interference and ensure the integrity of original wind speed data. A deep network is constructed by stacking multiple SDRSUs to extract useful features from noisy data. Considering the spatio-temporal coupling relationship between wind turbines in a wind farm, a ST-SDRSN (soft-activation based deep spatio-temporal residual shrinkage network) will be used to model the wind speed series neighboring time property and daily periodic property. An accurate wind speed prediction can be achieved by extracting the spatial correlations between the turbines at each turbine along the time axis. We designed four depth models under the same spatio-temporal architecture to verify the advantages of the soft-activation block and the proposed ST-SDRSN model. Two datasets provided by the National Renewable Energy Laboratory (NREL) were used for our experiments. Based on different kinds of evaluation criteria in different datasets, ST-SDRSN was shown to improve prediction accuracy by 15.87%.