This article is a part of a series of articles. You can find links to previous parts below:

Part I

Artificial bee colony algorithm

We do not use the term “busy bee” for no apparent reason. Bees are hard-working insects but are also very methodical. Bees in a bee hive are categorized into three broad categories: Scouts, onlookers, and employed bees. As shown in Figure 1, scouts are the bees that keep looking for new food sources. If they find a profitable food source (worth the effort, depending on how much nectar may be available), they will return to the hive and communicate the find via a dance. They then return to the food source with onlookers, who tag along to understand and map the location of these food sources. Then, these onlookers start foraging these food sources (collecting nectars), becoming employed. As you can imagine now, at any point, bees are, therefore, in a continuous process of exploration and exploitation.

Figure 1: Bee colony operation

Artificial bee algorithms leverage this concept of exploration and exploitation and the approach of counting and foraging. ABC algorithms are optimization-based algorithms used to solve a category of problems known as set-covering problems. The generalized problem statement of the set covering problem is that given a collection of elements, the solution aims to find the minimum number of sets that cover all of these elements.

So, in the context of ABC algorithms, within a particular space, there is a collection of food sources (a parameter in lieu of an actual food source, of course).

These sources, or food positions, are modified by the artificial bees (agents in the algorithm) with time to discover places of food sources with high nectar amounts and, eventually, the source with the highest nectar. Artificial bees fly (well, as you can assume, not literally) around in a multidimensional search space. Employed and onlooker bees choose food sources depending on their experiences and those of their nest buddies and adjust their positions accordingly.

Scouts, however, fly and choose the food sources randomly without using any previous experience. Suppose a new source’s nectar quantity (whatever parameter is in the model) is that of the previous one. In that case, it is higher than scouts memorizing the new position and forgetting the previous one. Hence, the ABC system combines local search methods (via employed and onlooker bees) with global search methods (via onlookers and scouts) and seeks to balance the exploration and exploitation process.

One of the best applications of this set of algorithms is in scheduling. These algorithms have been tried in many different scheduling scenarios, examples of which are in the papers linked below.

Paper 1

Paper 2

Paper 3

Paper 4

However, you have to remember that these algorithms tend to be leveraged in finding solutions to problems known as NP-hard. So, while they may sometimes yield better results than the previous approach, the results may be significantly different. So, while theoretically, they may be leveraged in labor and manufacturing job scheduling, they may not always be a significant improvement over other algorithms.

However, there is another approach to leveraging this algorithm for overall optimization. Like most examples I will discuss in this article series, executing this will need an architecture designed to cater to these or distributed AI algorithms. However, I believe that eventually, organizations will gravitate towards that type of architecture as computing power and technology mature to implement these algorithms quickly. However, coming back to the application of ABC in a different way, we can leverage the local and global approach to optimize tactical planning across a network.

The network, manufacturing, or warehousing must work in unison in a near real-time dynamic tactical planning world. That means the scheduling must be locally and globally (network-wide) optimal. And this is where algorithms like ABC can be helpful. However, the other caveat is that using optimization algorithms for real-time planning is challenging, considering the run time, specifically for NP-hard problems. This is where they can be combined with deep learning algorithms. Deep learning algorithms trained on ABC algorithms data can help perform real-time tactical planning.

We will keep going through the list of algorithms listed in Part 1 of the article in subsequent article. The third part of the article will be published on 11/29.