Algorithm Library for Medium and Long Term Electricity Sales Prediction Based on Adaptive Neural Network Learning

In recent years, with the continuous growth of China's electricity consumption scale, the impact of electricity consumption differences and uncertainty fluctuations in different industries on grid dispatching and safe operation continues to deepen. From the perspective of electricity sales, accurate prediction of electricity sales in different industries is one of the important means to solve the above problems. Therefore, this article proposes a medium and long-term prediction algorithm library that covers multiple industry categories and considers multiple time dimension characteristics. First, a deep neural network model is established to mine the characteristics of electricity sales under the annual, quarterly and monthly dimensions, fully considering the volatility and differences of electricity sales of different user groups. Then, a variety of medium and long-term prediction algorithm models based on temporal attention mechanism are established to achieve refined prediction of electricity sales based on different characteristics of change laws. Finally, an adaptive training mechanism is established for optimization model parameters, forming a refined model parameter automatic tuning strategy, achieving intelligent prediction model matching and accurate prediction of different characteristics of electricity sales. The proposed method has been verified to be able to achieve accurate prediction of electricity sales in multiple time dimensions and at multiple user scales.