Research On Multinodal Short-term Power Load Forecasting Based On Deep Neural Network Ensemble

The load forecasting module is very important in the daily operation,and highprecision load forecasting is a prerequisite to ensure the normal operation of the power system.In a large-scale industrial energy network,to achieve accurate load control,it is first necessary to accurately predict the load changes of multiple different key nodes.However,due to the characteristics of the production process and equipment operating conditions,the industrial load exhibits typical shock load characteristics.Strong volatility brings challenges to the accurate prediction of ultra-short-term power load.In response to this problem,this paper uses load factors,deep time series neural networks,and other methods to carry out short-term industrial power load forecasting research,and proposes a multinodal ultra-short-term load prediction model based on load factor,attention mechanism,and ensemble learning,deep time series neural networks,and integrated learning.To effectively utilize the timing characteristics between the load data,the Gated Recurrent Unit(GRU)is used as the base learner of the predictive model.Through the use of load factors to mine the coupling characteristics between the total node and the sub-nodes,extract the deep features of the load data of multiple nodes,add the attention mechanism to enhance the influence weight of the input features,and help the model make accurate decisions.The use of Bootstrap sampling and weighted average Hybrid integration strategy reduces errors caused by a single model and enhances the overall robustness of the model.Finally,through the use of the New Zealand open data set and the real collected data set for experimental comparison.Through the presentation of experimental data,the three evaluation indicators of MAPE,MAE,and NRMSE are compared with the results of multiple classic time series models and ensemble algorithms.The model in this paper has high precision and stable prediction results.