Abstract:
The current practise of pond operation of Upper Kotmale Hydropower Station is studied,
where management of the pond is by subjective judgements of the operator. Accurate and
reliable inflow forecast makes up an important basis for optimum pond operation connected
with effective spillway gate operation. This research proposes a novel technique to forecast
inflow to the pond and utilise these forecasts to optimise the operation of the pond.
In the first phase of the research, an artificial neural network based Nonlinear Autoregressive
eXogenous model, which is a dynamic neural network meant for time series forecasting, is
used to develop the real time inflow forecasting system. Cross correlation analysis is used as
feature selection for effective selection of the inputs to the Nonlinear Autoregressive
eXogenous network. In the second phase, real time inflow forecast for next six hours is used
to optimise the pond operation focusing on goals of shorter-term nature, such as maximising
power generation, maximising pond storage and minimising spillway discharge. Multiobjective
global optimisation using MATLAB “fmincon” algorithm and weighted approach
of solving multi-objective problem are utilised to solve the optimisation problem. Trading-off
conflicting objectives by this approach proves very effective. This optimisation approach
enhances the flexibility of the operator in the decision making process resulting in achievement
of efficiency in pond operation.
The results show that the Nonlinear Autoregressive eXogenous modelling is an efficient tool
for inflow forecasting and MATLAB “fmincon” algorithm can be used effectively to carry out
the multi-objective optimisation of run-of-river pond. Simulation studies for the past years
show that there exists an opportunity for optimising run-of river ponds for generation using
inflow forecast and with the use of the proposed methodology, it enhances the hydropower
generation with gains of over 5% which is significant in a plant of this type.
