Smart Grid Architecture ? Learning from the Human Brain

Authors:W. Wimmer, ABB Switzerland, N. R. Z├╝rcher, Center for Biomedical Imaging, Harvard Medical School, Charlestown, USA, R. D. Wimmer, New York University Langone Medical Center, USA

The question we have to ask ourselves is which of the newly possible architectures are reasonable and what can be gained through them in terms of availability and smartness? One approach to this is an analysis of different functional architectures for different purposes embedded in distinct system HW architectures, which determine availability and performance. Unfortunately the number of possible purpose/function combinations as well as physical architectures is rather large, and not all future purposes can be known in advance. To limit the breadth of our investigation, we may be able to get ideas from the features of the brain, which over the course of million years of evolution has successfully established adaptive networks.

At a first glance, microprocessor-based automation systems may appear completely different from neural systems, at least at the physical level. However, it has previously been reported that the brain also resembles a ‘small world’ architecture, both at the anatomical as well as at the functional level (Sporns 2004, Bassett 2006, Achard 2006, He 2007). Specialized neurons form local clusters with dedicated functionality as well as ‘long distance’ connections between different neuronal clusters. The main focus of the discussion that follows is at the functional level, i.e. physical architectures are put into second place. It is important to note that any availability or robustness lost at the functional level cannot be regained at the physical level except by physical redundancy - that is, unless the whole system is cloned.

Protection Functions
The most important function of living organisms apart from replication/reproduction is survival and therefore the protection against dangerous influences from the outside world. Likewise, protection of the process from outside influences is also the basic function of each process automation and protection system. Organisms as simple as bacteria, lacking the hallmarks of a brain, have evolved fast basic protection mechanisms, such as stimulus-response action schemes, which work similarly to the ones used in electromechanical protection relays. In animals with a nervous system, one of the behaviors that has been studied in depth is the withdrawal reflex of aplysia (Kandel 2000). This gastropod mollusk breathes through the gill, a delicate respiratory organ. In response to a tactile stimulus, the gill is rapidly withdrawn into a mantel cavity. A simple nervous system containing a short chain of sensory neurons, interneurons and motor neurons connects the mechanical receptors (sensory input) to the motor response (output) (Figure 2).

Those elements are comparable to the first numerical relays. In the aplysia, this setup enables basic learning behavior such as for example a decreased response to repeated stimulation (habituation) or an enhanced response to a novel (strong or noxious) stimulus (sensitization). Similar mechanisms still exist in humans and provide fast reflexes mediated via the spinal cord. For example when humans inadvertently put their hand on a heat source, they will reflexively and immediately withdraw it to avoid being burnt. While reflexes are relatively stereotyped behaviors, the development of a brain structure (as opposed to a loose nervous system) provides even more selective and complex responses in potentially dangerous situations. Biological systems have sensors and effectors that are separate from the brain (Figure 1). In a similar fashion, the HW architecture of numerical relays separates inputs from outputs and both together from the information processing microprocessor.

A psychological concept associated with self-preservation is the concept of ‘fear’. Fear is typically combined with an action that helps resolve a dangerous situation. Figure 3 shows the basic brain functions involved in fear processing, which typically starts with sensory input, followed by processing in the thalamus and the amygdala. The functional similarity with protection systems lies in the fact that nearly all standard protection functions such as those in the IEC 61850 data model provide a start signal alerting that a dangerous situation may be occurring.

Sensory input is preprocessed in the thalamus, which is connected with a wide range of brain regions. Joseph LeDoux showed the existence of both a fast subcortical as well as a slow cortical processing route (LeDoux 1994). The fast subcortical route conveys information from the thalamus directly to the amygdala, while the slower route transfers information to the cortex for further and more detailed processing before reaching the amygdala. Based on the direct thalamic input, the amygdala can trigger a fast physiological response to a possible threat via the emotional response control systems. In parallel, the slow route further evaluates sensory input and contextual information in higher-order cortical regions and the hippocampus. Both cortical areas and the hippocampus then send refined information to the amygdala to confirm or negate its initial response.

How does this compare to classical protection implementation? The sensory inputs we have for object protection are typically currents and voltages. The sensory processing function of the thalamus roughly corresponds to appropriate preprocessing and filtering of the analog inputs (logical nodes TCTR and TVTR in IEC 61850 terms). The protection functions have built-in (implicit) knowledge about dangerous situations, which based on the settings in non-volatile memory decide to set the start (action preparation) and possibly trigger an action. The trip conditioner (PTRC in IEC 61850 terms) then activates the ‘effective’ starting point declaring an emergency situation and issues the final trip (Figure 4). This classical protection implementation equivalent to the brain is shown in Figure 5. The protection functions together with PTRC decide to trip, which is similar to the brain triggering a reaction. However, the protection architecture is different from the one in the brain, because the fine-grained cortical processing does not only decide whether a final action is needed, but also which kind of action it should be. As object protection functions have no other choice than isolating the endangered object, this does not seem to be so crucial.

There are variations in system architectures. First, there are protection implementations where the alerting (amygdala-like) protection functions are different from the tripping (hypothalamus/pituitary-like) protection functions. For example an overcurrent protection function may detect a fault condition, but the tripping decision will be based also on the direction of the fault if such operating mode has been selected (Figure 4). Further, an automatic reclose at line protection might be considered as equivalent to a sensory cortex function reversing the previous trip decision (not shown in Figure 5).

Moreover, potentially different contexts as assessed by higher cognitive areas and the hippocampus are typically not used in a protection function. However, there are conditions in which contextual information may be desired. For example, if we think of seasons or weather conditions as ‘contexts’, there are situations where these may need to be considered, e.g. for the parameterization of the line protection function. In those circumstances, appropriate measurement information could help influence the protection function as to which set of parameter values from the memory settings it should use for decision-making. If this kind of function is used at all in today’s systems, it is mostly implemented by switching the active parameter value set from some external function e.g. at SCADA level. This kind of implementation however cannot act as quickly and specifically in response to changing weather conditions at a line as the human brain can. Thus, the architecture of typical object based protection functions is similar but not quite as flexible as the fear processing in the human brain. Future (or even a few existing, special) protection functions may use the extended brain architecture to create higher flexibility, while maintaining (or even enhancing) performance, e.g. by using contextual information like temperature to decide which set of protection parameters to select.

Power. Flexible. Easergy.
Protecting your electrical assets? today and tomorrow