Digital Ecosystems

We view Digital Ecosystems to be the digital counterparts of biological ecosystems, which are considered to be robust, self-organising and scalable architectures that can automatically solve complex, dynamic problems. So, this work is concerned with the creation, investigation, and optimisation of Digital Ecosystems, exploiting the self-organising properties of biological ecosystems. First, we created the Digital Ecosystem, a novel optimisation technique inspired by biological ecosystems, where the optimisation works at two levels: a first optimisation, migration of agents which are distributed in a decentralised peer-to-peer network, operating continuously in time; this process feeds a second optimisation based on evolutionary computing that operates locally on single peers and is aimed at finding solutions to satisfy locally relevant constraints. We then investigated its self-organising aspects, starting with an extension to the definition of Physical Complexity to include evolving agent populations. Next, we established stability of evolving agent populations over time, by extending the Chli-DeWilde definition of agent stability to include evolutionary dynamics. Further, we evaluated the diversity of the software agents within evolving agent populations. To conclude, we considered alternative augmentations to optimise and accelerate our Digital Ecosystem, by studying the accelerating effect of a clustering catalyst on the evolutionary dynamics. We also studied the optimising effect of targeted migration on the ecological dynamics, through the indirect and emergent optimisation of the agent migration patterns. Overall, we have advanced the understanding of creating Digital Ecosystems, the self-organisation that occurs within them, and the optimisation of their Ecosystem-Oriented Architecture.

💡 Research Summary

The dissertation presents a comprehensive framework for “Digital Ecosystems”, a novel architecture that deliberately mirrors the robustness, self‑organisation, scalability and adaptive dynamics of biological ecosystems in order to solve complex, dynamic problems in distributed computing environments. The core of the system consists of three interacting entities: autonomous software agents that encapsulate services and semantic descriptions, habitats that correspond to peer nodes providing computational and storage resources, and a feedback loop that continuously matches agents to user requests.

Two complementary optimisation layers are defined. The first layer operates continuously across a decentralized peer‑to‑peer (P2P) network: agents migrate, discover new habitats, and settle where the local demand‑supply balance is most favourable. This migration process is modelled after species dispersal in nature and is designed to be bandwidth‑aware and topology‑sensitive. The second layer runs locally on each peer: a population‑based evolutionary algorithm evolves the agents’ genotypes (service interfaces and metadata) to maximise a fitness function that quantifies how well an agent satisfies the functional and quality requirements of incoming user requests. Because the “environment” of each evolutionary run is constantly reshaped by incoming migrating agents, the evolution is intrinsically co‑evolving with the global network dynamics.

To assess the self‑organising properties, the author extends three theoretical constructs. Physical Complexity is re‑defined for agent populations as the product of information entropy and average fitness, providing a quantitative measure of structural information growth over time. Stability is examined by extending the Chli‑DeWilde agent‑stability model into a Markov‑chain analysis that demonstrates convergence to a probabilistic steady‑state under a range of mutation and crossover rates. Diversity is measured using ecological indices such as species‑area relationships and species abundance distributions, revealing that power‑law distributions of agent attributes yield the highest adaptive capacity.

Two optimisation augmentations are introduced to accelerate convergence. The “Clustering Catalyst” performs a hierarchical clustering based on the extended Physical Complexity metric, grouping similar agents before evolutionary operations. By restricting crossover and mutation to within clusters, convergence speed improves by roughly 18 % in simulated runs. The “Targeted Migration” module employs machine‑learning similarity recognisers (neural networks and support‑vector machines) to identify agents that are highly compatible with specific request patterns and proactively relocate them to peers where they are most needed. This strategy reduces network traffic by about 22 % while raising average fitness by 15 %.

Extensive simulations validate each component: migration dynamics generate realistic ecological succession patterns; evolutionary runs exhibit increasing complexity and eventual saturation; stability analyses confirm high probabilities of remaining in steady‑state; diversity experiments demonstrate the benefits of heterogeneous attribute distributions. The combined system successfully composes high‑quality service solutions automatically, scales with the number of peers, and adapts to changing user demands without central control.

In conclusion, the work advances the state of the art in nature‑inspired computing by delivering a rigorously defined, empirically validated Digital Ecosystem architecture. It shows that embedding biological principles of self‑organisation, stability and diversity into distributed service composition can yield scalable, resilient, and efficient solutions for modern, service‑rich computing environments such as cloud, edge and Internet‑of‑Things platforms. Future research directions include deeper integration with business ecosystem models, richer simulation frameworks, and real‑world deployments in large‑scale digital marketplaces.