Assessing the impact of weather-induced uncertainties in large-scale electricity systems

Abstract

The future energy system will largely depend on volatile renewable energy sources and temperature-dependent loads, which makes the weather a central influencing factor. This article presents a novel approach for simulating weather scenarios for robust large-scale power system analysis. By applying different signal analysis methods, historical weather data is decomposed into its spectral components, processed appropriately, and then used to generate random, self-consistent weather data. In this process, any weather parameters of different locations can be considered, while their respective dependencies are mapped. The added value is demonstrated by coupling with a state-of-the-art large-scale energy system model for Europe. It is shown that the integrated consideration of different weather influences allows a quantification of the range of fluctuation of various parameters - such as the feed-in of wind and solar power - and thus provides the basis for future resilient grid planning approaches.

Figures (8)

• Figure 1: Overview of the proposed model. Multiple time series of weather parameters of different locations are imported and preprocessed. This is followed by a regional aggregation of similar time series. In the Weather Model, a transformation to the frequency domain is carried out, which is used to calibrate the model and to generate random time series. These are used in the Energy System Model, which includes a convergence check to obtain the required amount of random weather years. The results comprise the effect of the weather on the distribution of all parameters that are considered within the energy system model, such as RES generation and power plant operation.
• Figure 2: Autocorrelation of temperature data in the time and frequency domain with 95 % confidence bounds. The correlation of neighbouring time steps is evident over a long period of time, while the dependence of adjacent frequencies is significantly lower, showing the benefit of analysis in the frequency domain.
• Figure 3: Absolute value of Fourier coefficients of three locations. The correlation between close locations 1 and 2 is higher compared to remote location 3. Furthermore, the correlation and the intensity decrease with a lower period duration. The peak at a duration of 24 h is an effect of the daily fluctuation, which is present in all time series.
• Figure 4: European noise intensity of different weather data. Regions with higher intensity tend to have higher fluctuations and are, therefore, more affected by increased future weather uncertainties. Furthermore, the Fourier transformation determines a plausible relationship between local phase angles, which represent the different times of sunrise.
• Figure 5: Left: Distribution of randomly generated hourly temperature data for an exemplary time period of 96 hours during summer. The orange area represents the 0.01 - 0.99 percentile range of all 947 randomly generated weather years to achieve convergence and the blue line the real temperature curve of the year 2012. Right: Eight randomly picked generated temperature data samples of the same period. This allows the observation of the temporal dependency structure of individual trajectories, while this is not possible with the marginal distributions on the left side.