Medium- and Long-Term Power Load Forecasting Method Based on LTC-RNN Model

Deng Bin, Zhang Nan, Wang Jiang, Ge Leijiao

Tianjin University, School of Electrical Automation and Information Engineering, Tianjin 300072

Deng Bin, Zhang Nan, Wang Jiang, et al. Medium- and Long-Term Power Load Forecasting Method Based on LTC-RNN Model [J]. Journal of Tianjin University (Science and Technology), 2022, 55(10): 1026-1133.

Deng Bin, Zhang Nan, Wang Jiang, et al. Medium- and Long-Term Power Load Forecasting Method Based on LTC-RNN Model [J]. Journal of Tianjin University (Science and Technology), 2022, 55(10): 1026-1133.

In response to the challenge of low accuracy in medium- and long-term power load forecasting affected by multiple uncertain factors such as weather, temperature, and holidays, this paper proposes a method based on a liquid time constant recursive neural network to achieve accurate forecasting of medium- and long-term power loads, laying a solid foundation for power system planning and design. The liquid time constant recursive neural network improves the state equation of neurons using membrane integrators and employs a semi-implicit Euler algorithm to complete the iterative update of neuron states, addressing the weak self-correction ability and the tendency to fall into local optima in recursive neural networks. This paper uses external influencing factors such as weather, temperature, and holidays as input to construct a medium- and long-term power load forecasting model based on a recursive neural network, and obtains the input-output mapping relationship through extended calculations in the hidden layers, achieving model training through the backpropagation algorithm over time to optimize the parameters of the neural network. In experiments, using actual grid data from a northern region as an example, the accuracy and efficiency of this forecasting method are validated. The research results show that the prediction accuracy of the liquid time constant recursive neural network for medium- and long-term power loads can reach 95.3%. Compared to Long Short-Term Memory (LSTM) and continuous-time recursive neural networks, this model exhibits more stable training results and higher prediction accuracy.

[Keywords]: Medium- and Long-Term Power Load Forecasting; Recursive Neural Network; Power System

Click the lower left corner “Read Original” to view the content of this issue