On May 22th, 23rd and 24th we attended the fourth International Conference on Future Energy Systems (ACM e-Energy) in Berkeley, California.
The conference was open by the beautiful keynote talk held by Dr. Jeffrey Taft from Cisco. It gave to the audience a complete understanding of current trends in the smart grid.
In their vision, the smart grid is a network combining 4 layers, the energy production and distribution, the market, the information layer and the social network (yes social networks). In the last years the progressive installation of renewable energy generators made the grid more and more unstable, as consequence of the sources exploited. To realize the vision of smart grid, it is necessary to provide the means for applications rather than looking for a killer application that justifies the need of such a change. Indeed innovation in the greed should foster developers, which would create a positive feedback to promote more and more innovation on the system. Basically, the stakeholders are looking for a system that can let them implement closed-loop controllers and attach them to the grid in a plug&play manner. This raises various concerns. The presenter mentioned data interoperability issues in the grid and showed the need of a multilayer network architecture able to tackle the complexity of the system.
In fact there is ongoing research in this direction (e.g. CoAP) but current solutions lack in scalability and still require high degree of mantainability. A multilayer network architecture should be deployed to give stakeholders the possibility to work at different abstraction levels and implement controllers as optimization problems broke down from the overall goal of ensuring a certain degree of reliability and stability.
In this model, the lower levels consist of components exploiting real-time delivery of information to manage safety-critical aspects. In the upper levels, the information is processed (e.g. to produce analytics) and delayed so that decision makers (both human and automatic) can exploit it to optimize specific goal functions.
In their vision, the smart grid is a network combining 4 layers, the energy production and distribution, the market, the information layer and the social network (yes social networks). In the last years the progressive installation of renewable energy generators made the grid more and more unstable, as consequence of the sources exploited. To realize the vision of smart grid, it is necessary to provide the means for applications rather than looking for a killer application that justifies the need of such a change. Indeed innovation in the greed should foster developers, which would create a positive feedback to promote more and more innovation on the system. Basically, the stakeholders are looking for a system that can let them implement closed-loop controllers and attach them to the grid in a plug&play manner. This raises various concerns. The presenter mentioned data interoperability issues in the grid and showed the need of a multilayer network architecture able to tackle the complexity of the system.
In fact there is ongoing research in this direction (e.g. CoAP) but current solutions lack in scalability and still require high degree of mantainability. A multilayer network architecture should be deployed to give stakeholders the possibility to work at different abstraction levels and implement controllers as optimization problems broke down from the overall goal of ensuring a certain degree of reliability and stability.
In this model, the lower levels consist of components exploiting real-time delivery of information to manage safety-critical aspects. In the upper levels, the information is processed (e.g. to produce analytics) and delayed so that decision makers (both human and automatic) can exploit it to optimize specific goal functions.
The other two keynote talks seemed to converge to the need of a distribution network closer to the internet. However, the existing infrastructure will unlikely be replaced with a new smart grid. We will most likely assist to the progressive replacement of crucial components with new ones. This will certainly bring high risk of failure into the system.
On the other hand, developing countries will be able to directly install innovative components into their energy systems, which will let them experiment new technologies and enable them to catch developed countries up. In the third keynote held by Arun Majumdar, the similarities with the internet network are even more evident. He clearly mention the need for something like power routers that would transform a strictly hierarchical network to a more flexible one. Current challenges and trends are very well presented by the speaker, altough for obvious reasons, ongoing solutions studied at Google are not even introduced to the audience.
Sessions:
- Storage integration
The first two papers dealt with the problem of mananging storage so as to reduce peak periods and optimize running costs of devices. The following two were as much interesting, as they proposed the use of batteries to reduce outages and improve power qualities in areas were energy provisioning is particularly unreliable. The third paper was in particular interesting, as it proposed to use a ups on a device-level, so that the ac-dc conversion losses can be minimized for consumer eletronics devices, and the problem of managing storage capabilities can be distributed. - Measurements and their use
The second session concerned the exploitation of consumption data to infer further information. The first paper presented an approach to demand forecasting, although they only used a feature for their predictor, which definetely makes the solution completely context-less. The second paper proposed a set of requirements and a cloud-based architecture very similar to the one proposed in our paper "Integrating households into the smart grid". However, they used a tailored protocol and data representation and did not consider scalability issues that might arise from this approach. The third paper presented a very interesting approach to infer household and inhabitants characteristics out of metering data. Finally the last paper proposed the use of ethnografy and other kind of studies used in humanities to enrich the data collected through sensor networks, so as to produce much more descriptive representations of inhabitants activities that can be better analyzed during studies promoting energy conservation. - Distributed control
The third session mostly included papers dealing with distributed vehicle charging. - Data center energy management
The papers of this session were related to energy-efficient data centres. In particular, the third paper dealt with the efficiency of ethernet interfaces. - Energy efficient networking and network inference
The first paper presented an approach to infer the topology of the distribution network out of time series collected on meters. The second paper introduced the concept of hierarchical state estimation in the energy grid and reported possible ways of detecting malicious attacks to this system. The following two papers dealt respectively with the performance of network interfaces and the TCP protocol. - Smart homes and buildingsThis session was definetely the last but not the least interesting. In the first paper an implementation of supervised random forest classifier was used to detect the device type connected to a sensing unit, as well as its operational state. In the third paper, a kinect interface was used to determine the amount of clothing weared by people. A voting mechanism was used as comfort feedback for the heating controller of the room. The final goal will be suggesting to the user to wear a coat when he start feeling cold and giving lower rates to the system. The last paper of the session concerned activity recognition in office spaces. In the study they used finite state machines to describe the state logic of the environment. However, this raised an interesting discussion on the need of domain expertise to specify the transitions. Therefore they decided to also use a LHMMs (Layered Hidden Markov Models) and learn the transition probabilities from a dataset. This model is then used to manage the light dimming and status. They showed remarkable savings of energy using this method.
More:
- Poster and Demo session
- The poster and demo session was no less interesting than the papers. Among those the work of University of Southampton should be mentioned. Basically, they use a HMM to build a model of inhabitants out of the events created in the household. Beside the HMMs they compared various algorithms, and showed HMM outperforming all of them. In fact it was very interesting to discuss the concept drift issue with classical data stream mining algorithms, as well as to the necessity of binding detected events to the inhabitant who actually performed it in order to improve the model.
- Panel discussionThe panel discussion was brought up a general discussion and interesting questions summarizing current trends and challenges in:
- Fossil fuels are not sustainable and we will have to cope with this problem already now, as climate change is evident.
- Renewable energy are destabilizing the energy grid, due to the volatile sources exploited for the production.
- Institution will play a crucial role: incentives will have drive customers to more efficient products and technologies, essential role is played by research and development activities.
- We will move to a situation where users are empowered with energy awareness and their decision making will be enriched by pervasive components, tracking their daily life activities and assisting them in reducing the footprint.
- Energy will have to flow in a bidirectional way and information regarding energy price will have to be available to users to get them into the loop.
