Neural network learns from mock-up operation experience : implementing on a solar energy community distribution

Lee, Chih-Hsiang and Csala, Denes (2020) Neural network learns from mock-up operation experience : implementing on a solar energy community distribution. Masters thesis, Lancaster University.

Text (2020Chih-Hsiangmres)
2020Chih_Hsiangmres.pdf - Published Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (4MB)

Abstract

Inspired by Imitation Learning, this dissertation trained a LSTM network by a mock-up operation experience of a solar energy community distribution system. Unlike the conventional method that implements LSTM only to predict features for the control programme to calculate an operation action according to a strategy, the LSTM of the proposed model integrates the strategy into its structure and thus can outputs actions directly. To examine whether the proposed model outperforms the conventional model, this dissertation first describes an operation strategy, adopted by both models, that aims to decrease total operation cost. Since the strategy needs accurate predictions to work effectively, an expert who can perfectly predict the future is created by historical data. The behaviours of the expert that follows the strategy are used as the training data of the LSTM in the proposed model. During simulation, the proposed model has better performance and computation efficiency than the conventional model by 25% higher and 75 times faster. Many researches have proposed control models for different systems and implemented LSTM only to predict key uncertainty in those models. To these researches, this dissertation demonstrates a promising result that the performance of a control model can be improved by integrating the strategy of that model into a neural network with mock-up operation experience.