Deep Neural Network-Based Optimisation for Clustered Demand-Side Energy Management in Smart Grids

This thesis conducts an in-depth investigation into the optimisation of Clustered Demand-Side Energy Management (CDEM) in smart grids through a Deep Neural Network (DNN)-based methodology. The study emphasises the strategic clustering of consumers and prosumers based on their energy consumption and generation profiles. By leveraging DNNs, the framework aims to accurately forecast energy demand, regulate load distribution, and support real-time optimisation objectives. The integration of Advanced Metering Infrastructure (AMI) and Internet of Things (IoT)-enabled devices underpins the data acquisition process, enabling precise clustering and dynamic energy management that is responsive to diverse user consumption behaviours.

The proposed architecture incorporates K-means clustering to group energy users according to their usage patterns. To enhance clustering performance, the study applies feature engineering techniques such as data scaling, correlation analysis, and comprehensive preprocessing. These measures contribute to more effective deployment of demand response strategies, ultimately reducing simultaneous peak demand and enhancing the operational stability of the smart grid. The DNN model is employed for high-fidelity prediction of energy usage and real-time demand response management. Its efficacy is validated through multiple performance metrics, including regression plots, confusion matrices, and error histograms, all of which reflect low prediction error, high classification accuracy, and consistent reliability across varied test cases. Real-world datasets obtained from smart meters further validate the model, offering granular insights into consumer energy usage dynamics.

Additionally, the research investigates voltage and frequency regulation using DNNs, comparing their performance to conventional PID controllers. The outcomes highlight the DNN model's superior accuracy and faster responsiveness in stabilizing grid operations. This thesis also explores fault and fraud detection mechanisms, applying DNNs to identify deviations in energy consumption patterns. The developed fault detection algorithms show notable improvements in detecting both supply-side and load-side anomalies. These algorithms are validated through Receiver Operating Characteristic (ROC) curve analysis, which measures the trade-off between true and false positives, and ensemble methods, further enhancing detection accuracy.

Linking to the clustered energy management framework, the research integrates renewable energy sources (RES) like wind and solar into these clusters, enabling optimisation of both conventional and renewable systems. Several case studies involving residential solar-powered systems and hybrid commercial loads have been implemented to evaluate the effectiveness of net-zero energy management strategies. These case studies highlight the critical role of battery energy storage systems (BESS) and demand response mechanisms in achieving optimal energy utilisation. The findings demonstrate the practical applicability of the proposed framework in diverse operational contexts, reinforcing its capability to support net-zero objectives through coordinated control of distributed energy resources. The results indicate significant improvements around 99.50% in energy efficiency and system reliability compared to traditional approaches, reducing energy waste and enhancing grid stability. Finally, the research provides key insights for future smart grid optimisation, emphasising the role of advanced AI techniques, such as DNNs, in advancing sustainable and resilient energy systems.

The key contributions of this research are as follows:
1. Novel Framework Design: Introducing a clustered-based DNN framework tailored for CDEM in smart grids.
2. DNN Implementations: Establishing the advantage of DNNs in handling large-scale, dynamic energy datasets through deeper architectures and enhanced learning capabilities.
3. Enhanced Energy Management: Demonstrating the effectiveness of clustering to capture consumer-specific energy usage patterns and optimize demand-side strategies.
4. Comprehensive Comparisons: Providing a detailed analysis of the proposed approach against existing methods, emphasizing its scalability, adaptability, and superior performance.

This work significantly advances smart grid technology, offering valuable insights for developing sustainable and efficient energy management systems. The findings underline the potential of DNN-based clustered frameworks