This PhD thesis presents a novel Kalman Filter Bank (KFB) methodology. The novel KFB is applied to predict and optimally schedule the operation of micro-grid energy resources with a public fixed-grid connection and an auxiliary battery storage backup facility. This methodology relies on a sliding single time step forecast projection to accurately acquire the anticipated schedule for a range of time dependent parameters, stochastic by nature, that have inherent embedded cyclical patterns. The innovative feature here consists of a novel bank of hourly adjacent Kalman Filter estimators, which operate in synchronism, for accurate parameter prediction for a selected time horizon. It is shown that, from consideration of the size (n) of the KFB framework (KFBn) the predictive performance of the algorithm can be greatly increased. For phenomena exhibiting a diurnal pattern a KFB₂₄ was employed with n ϵ [1,24]. For phenomena displaying an underlying weekly cyclical pattern a KFB₁₆₈ was employed for accurate prediction. Different case studies are investigated to substantiate the feasibility, accuracy and effectiveness of the proposed procedure with the primary focus on micro-grid renewable energy utilisation and optimised operation. For this purpose a fixed-grid connected university micro-grid, known as the Zero2020 test bed, based in Cork Institute of Technology (CIT), was used to develop and verify a novel predictive method to optimise the contribution of non dispatchable energy sources to the overall energy requirements of a college building. This thesis focuses on a holistic systems integration approach, whereby the micro-grid and energy storage element are integrated with the predictions of the future building energy demand into a single platform with future contribution of non dispatchable sources effected by external disturbances (weather). It was further employed to investigate the consequences of Time of Use (TOU) tariffs, which can be easily expanded to real time pricing (RTP), on the interactions between a 10kW wind turbine, a 12kW solar PV array, a 20 kWh Battery Power storage unit and the National Grid. A novel KFB₂₄ strategy was first applied to Numerical Weather Prediction (NWP) models to track the key characteristics of the variable wind speed, solar irradiance and ambient air temperature at fixed coarse grid node points to forecast the expected wind and solar PV power output for a college-based micro-grid electrical power network at an offset sub-grid point. The KFB₂₄ is then expanded to an novel KFB₁₆₈ for the purpose of mapping and deriving prediction for the weekly electrical load demand profile with varying college occupancy in scheduling electrical power resources. Key results, along with statistical hypothesis testing, are presented to validate and substantiate the claims made for the proposed KFB strategies and the accuracy of their prediction outputs. In an attempt to leverage the KFB method and further develop the concept of a single novel approach that can model the power output of the various renewable energy components of a micro-grid and also derive a day-ahead schedule of the energy exchange interplay between the battery and the national grid, the final step of this thesis focused on the expansion of the 24 KFB method into a novel holistic forecaster/optimiser tool. The thesis presents a novel KFB method, for forecasting RES over a 24 hour time horizon and to schedule the energy exchange within a fixed-grid connected college based micro-grid. This method relies on a developed Rule Based Logic Decision-making Processor (RBLDP) algorithm to control the power flow between the national grid, a battery storage system, the non-dispatchable energy resources and stochastic loads within the micro-grid in order to minimise the daily cost of energy import from the national grid based on time of use tariffs. Two different KFB frameworks are investigated as possible feasible holistic forecaster optimisation tools using the developed RBLDP algorithm. The first holistic KFB-RBLDP framework consisted of cascading a novel daily and weekly bank of 24 or 168 Kalman Filters, KFB₂₄ and KFB₁₆₈ respectively, operating in synchronism with a developed rule based optimised decision making strategy, to accurately predict the power contribution interplay from the micro-grid battery storage unit and the national grid with daily tariffs. This strategy is necessary in real time electricity markets where accurate day-ahead power prediction scheduling in crucial. Results are presented to demonstrate the accuracy of the predicted grid and battery power exchange when compared with the output of a simulated controller using measured imput data on a daily basis over several days. The variation of the recorded weekend loads compared with those during the normal working week is accurately captured and tracked by the 24-hour x 7 day extended Kalman Filter Banks (KFB₁₆₈) which maps the load behaviour of college occupancy. In addition, a 4 sigma statistical analysis, based on the Chebyshev inequality for measurements, is also presented to substantiate the accuracy of the prediction results obtained from the respective Kalman Filter Banks. A second KFB-RBLDP approach relies on the innate attributes of the Kalman Filter as a predictor to determine the optimised battery and grid power flow exchange. It was concluded that this approach yielded accurate results through direct application of filter iteration process for building electrical load demand with an innate periodicity based on a daily cycle. However, analysis of the former approach showed that it can capture weekend variations dissimilar to that during the normal working week. Furthermore, the predictive methods developed within this thesis are shown to have generalised time series prediction capabilities where the phenomena exhibits an underlying periodicity - (Abstract)

Thesis Cork Institute of Technology, 2017.

Bibliography: (pages 188-204)