Scheduling of Electric Energy in Smart Grids Using a Combination of Neural Networks and Local Optimization

In this paper a new scheduling algorithm is presented that enables fast calculation times with a combination of neural network and local optimization. Properly learned neural network is used to calculate schedule results that can be used as initial conditions for local optimization method. If scheduling results from neural network are located relatively close to optimal solution, local optimization method converges very fast and additionally improve scheduling result. To calculate results from neural network that are relatively close to optimal solution a learning dataset has to be obtained that contains optimal solutions. For this purpose genetic algorithms were used with large number of generations and several repetitions of scheduling process. Scheduling algorithm was tested on scheduling of the electric energy flows in smart grids to balance the electric energy consumption and production. Simulation results showed that a combination of neural network and local optimization converge faster than genetic algorithm method. This makes method useful where times for calculation of schedules are short.