Short-term Load Forecasting of Photovoltaic Distribution Grid Electricity Considering Carbon Emission Costs

In the load feature extraction stage, the traditional methods only consider the power load data itself, while ignoring other factors that may affect the load, resulting in the extracted features are not comprehensive enough, which leads to the low accuracy of the load prediction results. Therefore, a short-term load forecasting method for photovoltaic (PV) distribution network considering carbon emission cost is investigated. Historical load data of PV distribution networks are collected, and the data are filled with missing values, outliers are eliminated, data are normalized, and random noise is eliminated. Considering the influence of carbon emission cost, the Pearson correlation coefficient is used to screen the carbon emission cost features that are highly correlated with the historical load of photovoltaic distribution grid electricity, and the optimized feature set is combined with the convolutional neural network model and exploratory factor analysis method. Considering the characteristics of historical load data and the influence of carbon emission cost, the short-term load prediction model based on the long and short-term memory network is established, and the load features are extracted as inputs, and the short-term load prediction results of the photovoltaic distribution grid are outputted by the forward-propagation operation through the recurrent sub-network of the memory block of the model. The experimental results show that the research method can make the short-term load prediction curve of PV distribution network maintain high consistency with the actual curve, with high prediction accuracy, and the RMSE value is significantly better than that of the comparison method, which effectively improves the prediction accuracy.