Abstract
The effective prediction of bus load can provide an important basis for power system dispatching and planning and energy consumption to promote environmental sustainable development. A bus load forecasting method based on variational modal decomposition (VMD) and bidirectional long short-term memory (Bi-LSTM) network was proposed in this article. Firstly, the bus load series was decomposed into a group of relatively stable subsequence components by VMD to reduce the interaction between different trend information. Then, a time series prediction model based on Bi-LSTM was constructed for each sub sequence, and Bayesian theory was used to optimize the sub sequence-related hyperparameters and judge whether the sequence uses Bi-LSTM to improve the prediction accuracy of a single model. Finally, the bus load prediction value was obtained by superimposing the prediction results of each subsequence. The example results show that compared with the traditional prediction algorithm, the proposed method can better track the change trend of bus load, and has higher prediction accuracy and stability.
Highlights
Baojie HeWith the development of energy environment, the proportion of power consumption on the user side in energy consumption is gradually increasing
Variational modal decomposition (VMD) is an adaptive signal processing method proposed by Dragomiretskiy, which can be effectively applied to the smoothing processing of nonlinear and non-stationary time series [13]
Method is used to decompose, and each Intrinsic Mode Functions (IMF) component and residual component are obtained; (2) Normalize each sub-sequence component separately, and divide the training sample and the test sample according to the same ratio; (3) Construct an long and short-term memory (LSTM) neural network prediction model for each sub-sequence component, and use Bayesian optimization algorithm to optimize the hyperparameters of a single model to obtain the most suitable hyperparameter combination for decomposing the sequence and determine whether to use bidirectional long short-term memory (Bi-LSTM) in sub-sequences; (4) Train the prediction model after hyperparameter optimization, use the trained prediction model to perform multi-step extension prediction, and superimpose the reconstruction to obtain the multi-step prediction value of bus load; (5) Compared with actual data, the multi-step prediction performance of the prediction model is evaluated by calculating error indicators
Summary
With the development of energy environment, the proportion of power consumption on the user side in energy consumption is gradually increasing. Network (DNN) [1], Support Vector Machine (SVM) [2], long and short-term memory (LSTM) Network [3,4,5], DeepAR [6], N-BEATS [7], Transformer [8], etc These methods have the ability to model the nonlinear load process, can better adapt to nonlinear spikes and more accurately model the data characteristics of the load, and have better forecasting accuracy. The other is based on the nonlinearity of the bus load sequence non-stationary characteristics, using signal processing methods such as wavelet transform, EMD, and VMD to decompose the original sequence, separately model each sub-sequence component, and reconstruct its prediction results through superposition to obtain the combined prediction results that meet the accuracy requirements.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
