Data-Driven Modeling and Operation of Complex Energy Networks

Abstract

As global energy demand continues to rise and environmental issues become increasingly pressing, new energy sources are gradually replacing traditional fuels. However, integrating new energy sources into distribution networks presents many technical challenges, including effective load balancing, voltage control, and resource allocation. Traditional power flow calculations require knowledge of the distribution network's topology, but as more renewable energy sources and electric vehicles are connected to the grid, the network becomes increasingly complex. Sometimes, the topology of the network is unclear, and even when it is known, the parameters of the network may be difficult to obtain, making modeling challenging. Furthermore, when the topology is highly complex, it is difficult to establish a model using traditional methods. Therefore, this project proposes using a multilayer perceptron neural network to replace traditional power flow calculations and optimize distribution networks with renewable energy sources and intelligent electric vehicles. The objective is to achieve the lowest cost for the power grid while ensuring safe operation, without requiring knowledge of the network's topology. The project uses Pandapower to calculate power flows, trains MLPs to replicate traditional power flow calculations, and then uses the Big M method to transform battery segmentation function constraints and electricity price segmentation function constraints into linear programming problems, as well as to transform the maximum operator in the ReLU activation function of the MLP into a linear constraint with the Big M method. Finally, the entire optimization problem is converted into a mixedinteger linear programming problem for optimization, and the Gurobi solver is used to obtain the lowest cost while ensuring the safe operation of the power grid.