Bus load forecasting method based on DWT-SSA-Bi-LSTM neural network

Bus load forecasting is of great significance for the safe and stable operation of power grid and the balance of supply and demand. Due to the randomness, uncertainty and the impact of renewable energy, it is difficult to access high-precision prediction of bus load. Aiming at the problems existing in bus load forecasting, such as the non-stationary load curve, the lack of feature extraction, and the dependence on experience in hyper-parameter setting, this paper firstly used discrete wavelet to transform the bus load sequence, getting more periodic high-frequency and low-frequency components, then constructing the variant network of LSTM-Bidirectional LSTM. Sparrow search algorithm (SSA) is used to search the optimal hyper-parameters which contain the learning rate, the number of hidden neurons and the number of iterations. The final prediction results are obtained by predicting and re-constructing different components respectively. The experimental results are compared with SSA-Bi(Bi-bidirectional)-LSTM, Bi-LSTM, LSTM and BP neural network transformed by DWT(Discrete wavelet transform) and non-transformed SSA-Bi-LSTM, and it shows the prediction performance of DWT-SSA-Bi-LSTM proposed in this paper is better.