Perfect dispatch learning model based on adaptive Long Short-term Memory neural networks

Abstract

This paper proposes regarding economic dispatch problem as learning problem not optimization problem that develops adaptive Long Short-term Memory (LSTM) neural network to construct perfect dispatch learning model for each dispatch instant. First, matrix correlation analysis is performed to select critical historical instants for each dispatch instant. Then, according to matrix correlation analysis results, the structure of LSTM neural network at each instant is determined. Finally, the structures of LSTM learning models are adaptive for different dispatch instants. The differentiated training sets are constructed for adaptive LSTM learning models. The WSCC 9-bus system and the IEEE 118-bus system are used to verify the effectiveness of the proposed method. Study results indicate that the generation dispatch schedule obtained by the proposed LSTM learning model is closer to the perfect dispatch schedule than the generation dispatch schedule obtained by solving non-linear optimization problem.

Keywords

• Economic dispatch
• Machine learning
• Long Short-term Memory neural network