Medium-term Load Forecasting Method with Improved Deep Belief Network for Renewable Energy

With the continuous transition of traditional power system to newpower system, composition generation side in powersystem has gradually begun be dominated by renewable energy (at leastmore than 50%). Among sources, wind is themost susceptible weather and environmental influences. These factorsincrease complexity mode, put forwardhigher requirements for accuracy stability load forecasting. Thispaper proposes a medium-term forecasting methodbased on an improved deep belief network (IDBN-NN). The method includesthe construction network, layer-by-layer pre-trainingand fine-tuning model parameters, application model.In process parameter pre-training, Gauss-Bernoulli RestrictedBoltzmann Machine (GB-RBM) used as first part stacked deepbelief so that it can multiple types real-valued input data more effectively. In addition, IDBN-NN uses combination unsupervisedtraining supervised training pre-training. Finally, actual datais analyze calculation example. When number RBM layersis 3, fully connected layers 1, Dropout equal to0.2, MSE loss values are optimal, which 0.0037 0.0104,respectively. experimental results show proposed hashigher prediction when sample large loadinfluencing factors complex.