CONCLUSION AND FUTURE STEPS
We have presented a new method for adaptive anomaly detection and re-prediction based on pattern matching techniques.
We evaluated this method in the area of small scale residential electrical demand forecasting, a field of high interest
due to the emergence of important smart grid actors such as microgrids and VPPs.
The results obtained by our method are better than state of the art approaches in small scale, as shown in the results
section. Even more, as far as we know, this is the only forecasting approach in small scale that deals with normal
days as well as anomalous days without classification on a predetermined basis, thus enabling on-the-fly anomaly detection, pattern matching, and re-prediction techniques in case of unanticipated anomalous days occurring. We believe that the electrical demand forecasting results achieved are very good at residential transformer level, which in our case is considered to be of up to 350 kW.
We plan to improve our classification techniques in order for them to be based on seasonality and day of the week
patterns. For such improvements to occur, we need a larger dataset, one which spans several years, unlike our case which was limited to 17 months and therefore didn’t allow too much
tinkering in terms of SOM classes. Ultimately, we will connect our prediction techniques with demand response algorithms.
For this purpose our future work will test demand response multi-agent systems at household and community level based
on accurate power demand predictions, in order to suit the limits of the transformer providing power and also optimize
the use of available renewable sources. Furthermore, intelligent learning techniques involving collaboration should help in
fulfilling primary or critical objectives when combined to appropriate forecasting techniques. Some of our preliminary
tests already point out the benefits of such multi-agent systems in demand shifting [24]. Overall, these developments should all contribute to increased stability of the power system and a lower carbon footprint through efficient use of renewables, reduced user costs, and optimal operation of critical systems