Cruz E. Borges

Evaluating Combined Load Forecasting in Large Power Systems and Smart Grids

We present here a combined aggregative short-term load forecasting method for smart grids, a novel methodology that allows us to obtain a global prognosis by summing up the forecasts on the compounding individual loads. More accurately, we detail here three new approaches, namely bottom-up aggregation (with and without bias correction), top-down aggregation (with and without bias correction), and regressive aggregation. Further, we have devised an experiment to compare their results, evaluating them with two datasets of real data and showing the feasibility of aggregative forecast combinations for smart grids.

Efficient building load forecasting

The arrival of the smart grid paradigm has brought a number of novel initiatives that aim at increasing the level of energy efficiency of buildings such as smart metering or demand side management. Still, all of them demand an accurate load estimation. Short-term load forecasting in buildings presents additional requirements, among others the need of prediction models with simple or non-existing parametrisation processes. We extend a previous work that evaluated a number of algorithms to this end. Herewith we present several improvements including a variable data learning window and diverse learning data weighting combinations that further up improve our results. Finally, we have tested all the algorithms and modalities with four different datasets to show how the results hold up.

Short-term load forecasting in air-conditioned non-residential Buildings

Short-term load forecasting (STLF) has become an essential tool in the electricity sector. It has been classically object of vast research since energy load prediction is known to be non-linear. In a previous work, we focused on non-residential building STLF, an special case of STLF where weather has negligible influence on the load. Now we tackle more modern buildings in which the temperature does alter its energy consumption. This is, we address here fully-HVAC (Heating, Ventilating, and Air Conditioning) ones. Still, in this problem domain, the forecasting method selected must be simple, without tedious trial-and-error configuring or parametrising procedures, work with scarce (or any) training data and be able to predict an evolving demand curve. Following our preceding research, we have avoided the inherent non-linearity by using the work day schedule as day-type classifier. We have evaluated the most popular STLF systems in the literature, namely ARIMA (autoregressive integrated moving average) time series and Neural networks (NN), together with an Autoregressive Model (AR) time series and a Bayesian network (BN), concluding that the autoregressive time series outperforms its counterparts and suffices to fulfil the addressed requirements, even in a 6 day-ahead horizon.

Short-term load forecasting in non-residential Buildings

Short-term load forecasting (STLF) has become an essential tool in the electricity sector. It has been object of vast research since energy load is known to be non-linear and, therefore, very difficult to predict with accuracy. We focus here on non-residential building STLF, an special case of STLF where weather shows smaller influence on the load than in normal scenarios and forecast models, contrary to those on the literature, are required to be simple, avoiding dull and complicated trial-and-error parametrisation or setting-up processes. Under these premises, we have used a two-step methodology comprising a classification and a adjustment steps. Since the non-linearity of the load is associated to the activity in the building, we have demonstrated that the best way to deal with it is using the work day schedule as day-type classifier. Moreover, we have evaluated a number of statistical methods and Artificial Intelligence methods to adjust the typical hourly consumption curve, concluding that an autoregressive time series suffices to fulfil the requirements, even in a 5 day-ahead horizon.

Model selection in genetic programming

In this paper we discuss the problem of model selection in Genetic Programming. We present empirical comparisons between classical statistical methods (AIC, BIC) adapted to Genetic Programming and the Structural Risk Minimization method (SRM) based on Vapnik-Chervonenkis theory (VC), for symbolic regression problems with added noise. We also introduce a new model complexity measure for the SRM method that tries to measure the non-linearity of the model. The experimentation suggests practical advantages of using VC-based model selection with the new complexity measure, when using genetic training.

Smart buildings and the smart grid

So far the worlds of building automation and power networks have co-existed electrically attached but without data interaction, with different, and sometimes divergent, goals and requirements. The smartgrid community is coping nowadays with a new phase in the deployment of its vision: the integration of smart buildings, making use of their services and advantages to enable more complex functionalities at a global level. The main objective of this paper is to review latest results of the research community of industrial electronics, whose society within the IEEE, the IES, acts as the organizer of this conference. At the conference, latest advances and developments in design, modelling, simulating and implementing tools for, or systems of, sensor and/or actuator networks with advances towards user orientation, wireless connectivity, dependability, energy efficiency, context awareness and ubiquitous computing will be presented.

Penalty functions for genetic programming algorithms

Very often symbolic regression, as addressed in Genetic Programming (GP), is equivalent to approximate interpolation. This means that, in general, GP algorithms try to fit the sample as better as possible but no notion of generalization error is considered. As a consequence, overfitting, code-bloat and noisy data are problems which are not satisfactorily solved under this approach. Motivated by this situation we review the problem of Symbolic Regression under the perspective of Machine Learning, a well founded mathematical toolbox for predictive learning. We perform empirical comparisons between classical statistical methods (AIC and BIC) and methods based on Vapnik-Chrevonenkis (VC) theory for regression problems under genetic training. Empirical comparisons of the different methods suggest practical advantages of VC-based model selection. We conclude that VC theory provides methodological framework for complexity control in Genetic Programming even when its technical results seems not be directly applicable. As main practical advantage, precise penalty functions founded on the notion of generalization error are proposed for evolving GP-trees.

A NEW LINEAR GENETIC PROGRAMMING APPROACH BASED ON STRAIGHT LINE PROGRAMS: SOME THEORETICAL AND EXPERIMENTAL ASPECTS

Tree encodings of programs are well known for their representative power and are used very often in Genetic Programming. In this paper we experiment with a new data structure, named straight line program (slp), to represent computer programs. The main features of this structure are described, new recombination operators for GP related to slps are introduced and a study of the Vapnik-Chervonenkis dimension of families of slps is done. Experiments have been performed on symbolic regression problems. Results are encouraging and suggest that the GP approach based on slps consistently outperforms conventional GP based on tree structured representations.

On the probability distribution of data at points in real complete intersection varieties

We show several estimates on the probability distribution of some data at points in real complete intersection varieties: norms of real affine solutions, condition number of real solution of real systems of multi-variate polynomial equations and convergence radius of Newtons operator for under-determined system of multi-variate polynomial equations.

Adaptation, Performance and Vapnik-Chervonenkis Dimension of Straight Line Programs

We discuss here empirical comparation between model selection methods based on Linear Genetic Programming. Two statistical methods are compared: model selection based on Empirical Risk Minimization (ERM) and model selection based on Structural Risk Minimization (SRM). For this purpose we have identified the main components which determine the capacity of some linear structures as classifiers showing an upper bound for the Vapnik-Chervonenkis (VC) dimension of classes of programs representing linear code defined by arithmetic computations and sign tests. This upper bound is used to define a fitness based on VC regularization that performs significantly better than the fitness based on empirical risk.