Short-Term Net-Load Forecasting Based on LSTM Enhanced by Adaptive Adjustment and Multi Period Prediction Mechanism

As one of the most challenging issues in the new type power systems, many scholars pay attention to net-load forecasting. Aimed at this problem, a combined forecasting model based on long short term memory (LSTM) enhanced by adaptive adjustment and multi period prediction mechanism is presented. In this paper, Gaussian Copula correlation analysis is used for mining the relationship between net-load and distributed energy generation, with the advantage of complexity quantification. Then, the base model of net-load forecasting is built by LSTM due to its advantage of mining the useful information behind the historical samples. A polynomial decay learning rate adaptive adjustment method is introduced for improving the overall training effectiveness of LSTM. Moreover, multi period prediction mechanism can enhance the temporal coupling characteristics among the features. From the simulation experiments, the proposed methodology can perform best among all the algorithms for comparison.