Dalarna University's logo and link to the university's website

Solar energy is experiencing a rapid growth worldwide and it is a promising technology to play a vital role in future power systems. However, the natural intermittent of solar energy resources affect the quality of the output power from a solar system putting its dispatchability in question. Battery Energy Storage System (BESS) is among the solutions to suppress PV power fluctuations and allow smooth PV power production as result. In this work, BESS was modelled, integrated and simulated for Agahozo Shalom Youth Village (ASYV) PV Power Plant located in Rwanda. MATLAB/Simulink was used as modelling and simulation tool and Excel was used in data analysis. For ease of work, one of eight PV arrays at the power plant was considered and the results were scaled up to the entire power plant afterwards.

Annual data for year 2017 was analysed and the worst-case scenario which is the day with highest irradiance variation was found. Both PV power and ambient temperature data for this day were used as input to the model. Four smoothing approaches namely Low Pass Filter (LPF), Simple Moving Average (SMA), Exponential Moving Average (EMA) and Ramp Rate Control (RRC) were investigated in detail, applied to the model and compared in terms of performance and battery size that each approach would require. The degree to which power output needs to be smoothed can vary based on regulatory requirements and the technical conditions of the power grid. For this reason, three Ramp Rate Limits (RRLs) namely ±10 %, ±20 % and ±30 % of the rated PV array power per minute were applied in smoothing algorithms to see how large the battery storage would be if Rwandan grid operator was to impose one of the aforesaid RRLs.

The results showed that all smoothing methods managed to smoothen out PV array power at all RRLs as intended. The difference occurred in performance of smoothing methods and battery size in terms of power and energy that each method required. In all cases, it was noticed that both LPF and EMA displayed almost similar results which made it difficult to make a clear distinction between the two. However, in their slight difference, EMA required a slightly smaller battery size. The memory effect of SMA was noticed and this method was requiring bigger battery size at all RRLs. The RRC performance was better especially at ±10 % RRL compared to other three methods. The particularity of RRC was that it only allows the battery to respond when needed and the battery charges or discharges the exact amount of power needed. This was different from the other three smoothing methods. They were always allowing the battery to respond even when the present power ramp is within the set RRL resulting in high charging/discharging cycles which causes cyclic degradation of the battery. In addition, these methods over-smoothed the PV array power where more than needed power could be absorbed or delivered by the battery resulting in unnecessary bigger battery size.

Some of downsides of RRC method were that it requires bigger battery size in terms of energy and it could be more sensitive to the uncertainty associated with PV array power measurement compared to other methods. Nonetheless, its battery size in terms of power requirement was less than other methods since it does not over-smooth the PV array power. Since RRC and EMA methods were requiring less battery power and less battery energy at all RRLs respectively, both methods were chosen while scaling up the results to the entire power plant. Using EMA smoothing methods over RRC at ±10 % RRL could results in saving $113 thousand of battery capital cost. However, at ±20 % and ±30 % RRL, the RRC method was found to be the best option since it needs less capital cost than EMA smoothing approach.