Data Analytics in Power Engineering and Demand Management

Through this thesis, we attempt\ua0to bridge the two recent advancements in power engineering - Demand Management and Data Analytics in power engineering. While both these branches are stand alone, effective Demand Response can be facilitated only by improving the understanding of the demand and the customer behavior that contributes to electricity consumption. This thesis investigates these two aspects independently in its two parts and give exhaustive solutions for both.In the first part of the thesis, a procedure is presented to aid network operators and electricity service utilities make use of the available big data in the realm of power networks. To illustrate the procedure, one of the data intensive applications in power networks - customer energy profiling - is selected. While existing protocols utilize a limited number of variables to forecast the electricity consumption, partly due to the complexities that arise when dealing with large number of variables, our approach attempts to handle high dimensional data in the order of hundreds of variables. The variables are hinged on household specific characteristics that are available in the public domain and/or those could be fetched by the network operators with minimal effort. Should the power system operators/retailers be capable of obtaining this data from their customers through methods such as questionnaires and surveys, energy modelling can be performed accurately together with the available real-time data. This part provides an instructional layout detailing all the computational working codes. The tools developed are unique in the sense that to the best of authors' knowledge, there is no package available in Julia, the language we relied for code development that details all the data processing stages and model selection. The virtue of our model is validated with another popular statistical model selection method, Lasso. Details of our methodology and the working codes are provided through GitHub so that researchers in the community may adopt and/or build on this approach.In the second part of the thesis, a real-time control mechanism is devised to aid electricity network operators facilitate effective residential Demand Response. The novelty of our mechanism is that, contrary to the existing algorithms where individual customer behaviour patterns are required or estimated, here, a novel load management scheme is proposed for highly stochastic loads where each household possesses individually suited parameters. If the actual demand in a feeder is found to exceed its rating, selected customers are requested by the operator to lower their loads. The benefit of the proposed approach is to maximise the certainty of meeting Demand Response targets whilst minimising the overall cost associated with the Demand Response control actions. The contribution of this thesis part is a model that takes customer behaviour patterns into consideration while selecting customers. The load behaviour of individual households is modelled using Markov chains and by treating the problem as a Markov Decision Process optimal control problem, an easily implementable strategy for load reduction is arrived. This strategy is then incorporated in an aggregate model that utilises household specific transition probabilities.The practical application of the developed controller is substantiated based on real data. Moreover, the numerical results illustrate the virtue of the novel strategy, indicating potential for the use of household specific Markovian models for Demand Response applications