|
演讲摘要: The purpose of a Typical Meteorological 5-Year series (TM5Y) is to fit the daily meteorological variability as well as the long-term variability, assessed both monthly and annually.
A TM5Y series describes extreme realistic months and years that allow for the sizing of a PV system with storage. The monthly and annual variability of PV power production is also a key focus for grid with a high percentage of renewable energy sources and the financial model.
Context:
Accurate climate resource quantification is crucial in several different fields such as agriculture, wind and solar energies or net zero energy building.
Oftentimes, climatic information is fragmented and difficult to obtain in fine time steps. In the case a series of climatic data is complete over a long period of time, its processing and use in simulation tools require too long calculation time.
For these reasons, over the last forty years, the scientific community has been working on condensing meteorological data into a Typical Meteorological Year (TMY), a tool that characterises the typical daily distribution for each month of the year without erasing the impact of the fluctuations within one day (e.g. averages, extreme values, distributions, etc.).
State of the art:
There are as many different TMYs as there are statistical methods that can create them. In 1999, Agirou and al. compared no less than 17 of them. The Filkenstein-Schafer’s method is one of the most used nowadays. It is based on the comparison of cumulative distribution functions (CDF) and produces satisfactory output for the construction of a TMY. Among the other popular methods is that of Sandia Laboratories, which developed the standard TMY2S and TMY3S series built by the NREL. This method first defines 12 average values for each climatic parameter of a long Time Series (TS), one for each month. Then, it selects from the TS the month nearest to the corresponding average value to shape the hourly values of each month of the TMY.
Barriers:
Long TS and TM5Y are widely spread across the scientific literature, but no in-between solution currently exists. Nonetheless, some applications, such as storage systems, require one to know the monthly distribution and boundary values of a specific month (eg January) in order to forecast more extreme events. But the particular data required is not always accessible over long TS or is too costly.
Conclusion: This lack of intermediate-size climatic series must be fulfilled. Climatic series should accurately and succinctly describe the typical layout for long compact TS.
Work carried out:
The aim is to provide a short typical climatic series (5 years): TM5Y. Thus, the TMY monthly values encompass the TS’s extrema, averages and standard deviations. The method applied in this document uses monthly parameters to generate the hourly climatic series.
TM5Y methodology
• Similar to the Sandia Laboratories method, the first step is to sort the monthly values available in the TS in a year/month table.
• Then, still following Sandia Laboratories method, average for each month is calculated. Three other values are also taken into account per month: the standard deviation, the minimum and the maximum. Thus, four reference values are obtained for all 12 months. The same is done for the yearly figures.
• The third step involves selecting 5 months from 5 different years of the TS for each of the 12 months. This selection is done by minimising the weighted distance between the TS and the chosen months using the 4 reference values from step 2. This difference is calculated by applying a heavy weight to the average and a light weight to the extrema. The number possible combinations of 5 months out of Nmonth years where the data for that month is available is given by:
This number is low enough that all scenarios can be tested.
At this stage, the quality of the monthly information is superior to that of the classic TMY. The monthly distributions are closer to that of the complete TS and they are compatible with the energy storage system sizing.
The graph below compares the averages, standard deviations and extrema of the TS and the TM5Y sample. The weighting ratios have been carefully selected to minimise the difference between the two series of data.
• The final step is to use the previously selected months to build 5 typical years. Not all combinations of twelve months yield realistic years: if one year was constructed using only the worst months of the TS for each of the 12 months, that year would have a much lower sum than the observed minimum annual values of the TS. The number of possible combinations, i.e. , is too big to test each of them. A random screening method combined with our convergence algorithm enables construction of the target years. Finally, each year built is assigned its percentile according to the normal distribution of annual irradiation (P10, P27, etc.).
Conclusion:
The TM5Y fills the gap between the long Time Series and the Typical Meteorological Year. Its monthly representativeness satisfies the needs of the study of storage systems. Furthermore, the annual representativeness allows for sensitivity studies (financial, energy-related) over long time periods.
Financial and energy studies over long time-periods can now be undertaken thanks to TM5Y annual representativeness.
Future work
In this document, the suggested approach is based on the Sandia Laboratories method. A Filkenstein-Schafer approach based on cumulative distribution functions rather than monthly values should improve our results.
The convergence algorithm used to build the 5 typical years could be improved to ensure the convergence towards a global optimum in a small timeframe.
|