Astrid Zeman

A Simulator for Self-Adaptive Energy Demand Management

A Demand-Side Program Simulation Tool is designed to predict the response from different deployment strategies of distributed domestic energy management. To date, there are several case studies of demand management and control projects from around the world. To achieve results with sufficient generality, case studies need to be conducted over long periods, with a reasonable number of diverse households. Such case studies require large capital to set up hardware and software.To bypass these financial and temporal investments, we have designed a simulator for energy suppliers to use in order to learn the likely performance of large-scale deployments. Of main interest is the prediction of not only the level and firmness of demand response in critical peak pricing trials, but also the householdpsilas comfortable level and satisfaction level. As an example of the power of the simulator we have used it to develop and test a new self-adaptive methodology to intelligently control the energy demand. The methodology is adaptive to global factors, such as the market energy price, as well as local conditions, such as the satisfaction level of households. This paper outlines self-adaptive methodologies used within the simulator. Experimental results show energy consumption under different control strategies and the improvement of system behavior through adaptive design. With the self-adaptive demand management strategy, the total energy consumed by one million householdspsila controllable loads has reduced dramatically while the satisfaction level of households is well maintained. This is one of the very first simulators that take into account both technical and human behavior aspects.

Adaptive Control of Distributed Energy Management: A Comparative Study

Demand-side management is a technology for managing electricity demand at the point of use. Enabling devices to plan, manage and reduce their electricity consumption can relieve the network during peak demand periods. We look at a reinforcement learning approach to set a quota of electricity consumption for a network of devices. This strategy is compared with homeotaxis - a method which achieves coordination through minimizing the persistent time-loop error.These policies are analyzed with increasing levels of noise to represent loss of communication or interruption of device operability. Whilst the policy trained using reinforcement learning proves to be most successful in reducing cost, the homeotaxis method is more successful in reducing stress on devices and increasing stability.

The Müller-Lyer Illusion in a Computational Model of Biological Object Recognition

Studying illusions provides insight into the way the brain processes information. The Müller-Lyer Illusion (MLI) is a classical geometrical illusion of size, in which perceived line length is decreased by arrowheads and increased by arrowtails. Many theories have been put forward to explain the MLI, such as misapplied size constancy scaling, the statistics of image-source relationships and the filtering properties of signal processing in primary visual areas. Artificial models of the ventral visual processing stream allow us to isolate factors hypothesised to cause the illusion and test how these affect classification performance. We trained a feed-forward feature hierarchical model, HMAX, to perform a dual category line length judgment task (short versus long) with over 90% accuracy. We then tested the system in its ability to judge relative line lengths for images in a control set versus images that induce the MLI in humans. Results from the computational model show an overall illusory effect similar to that experienced by human subjects. No natural images were used for training, implying that misapplied size constancy and image-source statistics are not necessary factors for generating the illusion. A post-hoc analysis of response weights within a representative trained network ruled out the possibility that the illusion is caused by a reliance on information at low spatial frequencies. Our results suggest that the MLI can be produced using only feed-forward, neurophysiological connections.

A Reinforcement Learning Approach to Setting Multi-Objective Goals for Energy Demand Management

In order to cope with the unpredictability of the energy market and provide rapid response when supply is strained by demand, an emerging technology, called energy demand management, enables appliances to manage and defer their electricity consumption when price soars. Initial experiments with our multi-agent, power load management simulator, showed a marked reduction in energy consumption when price-based constraints were imposed on the system. However, these results also revealed an unforeseen, negative effect: that reducing consumption for a bounded time interval decreases system stability. The reason is that price-driven control synchronizes the energy consumption of individual agents. Hence price, alone, is an insufficient measure to define global goals in a power load management system. In this article we explore the effectiveness of a multi-objective, system-level goal which combines both price and system stability. We apply the commonly known reinforcement learning framework, enabling the energy distribution system to be both cost saving and stable.

Homeotaxis: Coordination with Persistent Time-Loops

We present a novel approach to self-organisation of coordinated behaviour among multiple resource-sharing agents. We consider a hierarchical multi-agent system comprising multiple energy-dependent agents split into local neighbourhoods, each with a dedicated controller, and a centralised coordinator dealing only with the controllers. The coordinated behaviour is required in order to achieve a balance between the overall resource consumption by the multi-agent collective and the stress on the community. Minimising the resource consumption increases the stress, while reducing the stress may lead to unrestricted and highly unpredictable demand, harming the individual agents in the long-run. We identify underlying forces in the systems dynamics, suggest a number of quantitative measures used to contrast different strategies, and introduce a novel strategy based on persistent sensorimotor time-loops: homeotaxis. Homeotaxis subsumes the homeokinetic principle, extending it both in terms of scope (multi-agent self-organisation) and the state-space, and allows to select the best adaptive strategy for the considered system.

Decentralized Multi-Agent Clustering in Scale-free Sensor Networks

Many interaction processes in complex adaptive systems occur in groups, and in order to organize knowledge, collaboration and a proper distribution of functions and tasks, there is a need to analyze, model and develop computational systems in which several autonomous units interact, adapt and work together in a common open environment, combining individual strategies into overall behavior. The approach to engineering a desired system-level behavior, adopted in this work, is based on a multi-agent system [11], in which the preferred responses emerge as a result of inter-agent interactions.

An exponential filter model predicts lightness illusions.

Lightness, or perceived reflectance of a surface, is influenced by surrounding context. This is demonstrated by the Simultaneous Contrast Illusion (SCI), where a gray patch is perceived lighter against a black background and vice versa. Conversely, assimilation is where the lightness of the target patch moves toward that of the bounding areas and can be demonstrated in Whites effect. Blakeslee and McCourt (1999) introduced an oriented difference-of-Gaussian (ODOG) model that is able to account for both contrast and assimilation in a number of lightness illusions and that has been subsequently improved using localized normalization techniques. We introduce a model inspired by image statistics that is based on a family of exponential filters, with kernels spanning across multiple sizes and shapes. We include an optional second stage of normalization based on contrast gain control. Our model was tested on a well-known set of lightness illusions that have previously been used to evaluate ODOG and its variants, and model lightness values were compared with typical human data. We investigate whether predictive success depends on filters of a particular size or shape and whether pooling information across filters can improve performance. The best single filter correctly predicted the direction of lightness effects for 21 out of 27 illusions. Combining two filters together increased the best performance to 23, with asymptotic performance at 24 for an arbitrarily large combination of filter outputs. While normalization improved prediction magnitudes, it only slightly improved overall scores in direction predictions. The prediction performance of 24 out of 27 illusions equals that of the best performing ODOG variant, with greater parsimony. Our model shows that V1-style orientation-selectivity is not necessary to account for lightness illusions and that a low-level model based on image statistics is able to account for a wide range of both contrast and assimilation effects.

Complex cells decrease errors for the Müller-Lyer illusion in a model of the visual ventral stream

To improve robustness in object recognition, many artificial visual systems imitate the way in which the human visual cortex encodes object information as a hierarchical set of features. These systems are usually evaluated in terms of their ability to accurately categorize well-defined, unambiguous objects and scenes. In the real world, however, not all objects and scenes are presented clearly, with well-defined labels and interpretations. Visual illusions demonstrate a disparity between perception and objective reality, allowing psychophysicists to methodically manipulate stimuli and study our interpretation of the environment. One prominent effect, the Müller-Lyer illusion, is demonstrated when the perceived length of a line is contracted (or expanded) by the addition of arrowheads (or arrow-tails) to its ends. HMAX, a benchmark object recognition system, consistently produces a bias when classifying Müller-Lyer images. HMAX is a hierarchical, artificial neural network that imitates the “simple” and “complex” cell layers found in the visual ventral stream. In this study, we perform two experiments to explore the Müller-Lyer illusion in HMAX, asking: (1) How do simple vs. complex cell operations within HMAX affect illusory bias and precision? (2) How does varying the position of the figures in the input image affect classification using HMAX? In our first experiment, we assessed classification after traversing each layer of HMAX and found that in general, kernel operations performed by simple cells increase bias and uncertainty while max-pooling operations executed by complex cells decrease bias and uncertainty. In our second experiment, we increased variation in the positions of figures in the input images that reduced bias and uncertainty in HMAX. Our findings suggest that the Müller-Lyer illusion is exacerbated by the vulnerability of simple cell operations to positional fluctuations, but ameliorated by the robustness of complex cell responses to such variance.

Symbiotic sensor networks in complex underwater terrains: a simulation framework

This paper presents a new multi-agent physics-based simulation framework (DISCOVERY), supporting experiments with self-organizing underwater sensor and actuator networks. DISCOVERY models mobile autonomous underwater vehicles, distributed sensor and actuator nodes, as well as multi-agent data-to-decision integration. The simulator is a real-time system using a discrete action model, fractal-based terrain modelling, with 3D visualization and an evaluation mode, allowing to compute various objective functions and metrics. The quantitative measures of multi-agent dynamics can be used as a feedback for evolving the agent behaviors. An evaluation of a simple simulated scenario with a heterogeneous team is also described.

Predicting cluster formation in decentralized sensor grids

This paper investigates cluster formation in decentralized sensor grids and focusses on predicting when the cluster formation converges to a stable configuration. The traffic volume of inter-agent communications is used, as the underlying time series, to construct a predictor of the convergence time. The predictor is based on the assumption that decentralized cluster formation creates multi-agent chaotic dynamics in the communication space, and estimates irregularity of the communication-volume time series during an initial transient interval. The new predictor, based on the auto-correlation function, is contrasted with the predictor based on the correlation entropy (generalized entropy rate). In terms of predictive power, the auto-correlation function is observed to outperform and be less sensitive to noise in the communication space than the correlation entropy. In addition, the preference of the auto-correlation function over the correlation entropy is found to depend on the synchronous message monitoring method.