In recent years, both rooftop distributed and large-scale centralized solar power generation have achieved rapid development. For grid operators, the ideal power supply should have the characteristics of high reliability, predictability, and on-demand supply, but sunlight exposure through the complex and changeable weather conditions between the sun and the ground makes the solar energy resources have high changes This poses a major challenge for the utilization of solar energy and grid operators. Therefore, the U.S. Department of Energy ambitiously launched a Sunshot program, trying to significantly reduce the cost of solar installation to enhance its competitiveness. One of the tasks of the plan is to find ways to reduce the overall cost, so that the price of solar power generation is equal to the grid, so as to increase its penetration rate in the market. Among them, ancillary services can significantly reduce overall costs. For example, reliable solar power generation forecasts can reduce equipment idling. The Western Wind and Solar Energy (WWSIS) study found that when solar and wind power generation accounted for 30%, the power generation forecast service can save the Western Power Coordination Commission (WECC) by 14% (for example, up to US$5 billion) in operating costs . In addition, California independent system operators (CallSO) recommended in a report to predict solar energy 1 hour in advance or within a different period of a day to reduce overall costs.
Ultra-short-term solar prediction (for example, 0~3h) requires accurate prediction of the time and space details of the downstream radiation field, including capturing high-frequency reflections in the radiation field due to cloud occlusion or aerosol attenuation, and local cloud layers/ The influence of aerosol distribution on scattered radiation in the sky. This is because cloud cover is the main cause of solar changes, and this is especially true in ultra-short-term forecasts.
Long-term predictions (a few hours to a few days) require accurate initialization of the model and display of real clouds. When evaluating resources, it takes longer time series (multi-year) observational data to achieve robust statistics, and finally obtain seasonal averages and characteristics. Among them, from atmospheric circulation to micro-meteorology affected by topography will affect the results of the above statistics. However, the development of climate prediction models is obviously better than that of the atmosphere and its weather sciences, which started late. Cloud attributes and their complex conditions in feedback, affect the current state of the climate and its natural variability (that is, the nature of cloud distribution and the cloud itself Content) to reduce the predictive ability.
The satellite observation system is an indispensable tool in the process of advancing the development of solar energy companies. It can meet the all-round needs of customers: including resource assessment at the climate level, and small enough to analyze the operation load balance prediction of a single cloud at the time and space scale.
From the processing capabilities of the different prediction methods shown in Figure 1(a), especially the content of cloud layer prediction in Figure 1(b), we can see the importance of satellite data in solar energy forecasting and resource assessment. Generally, the initial time and the period after the initial observation are more representative of the current irradiance than the model estimates. In a very short prediction time interval (<30min), applying simple linear assumptions to observations can well describe the current cloud field changes (growth/decay and motion). This “short-term prediction” has great certainty, and it also requires the observation system to provide the following: accurate cloud cover distribution (horizontal and vertical) information; estimate cloud movement based on attribute tracking; can describe the effect of clouds on direct beams and beams. (E.g. shadow) optical performance evaluation of the influence of scattered radiation (e.g. side scatter). At this time, regional observation equipment such as all-sky imagers can effectively provide the passage time of cloud shadows in a short period of time.
However, it may be difficult for an all-sky imager to predict the impact of clouds on a certain surface position within 20-30 minutes (also depending on the obstacle of the field of vision). At this time, satellite inversion resources can be used, especially the use of faster update satellites to expand the time range of the forecast. Based on the restrictive assumption that cloud properties (for example, no growth/attenuation and fixed optical properties) will not change, simple advection motion caused by wind direction derived from satellite remote sensing or numerical weather prediction (NWP) models can be advanced 1~3h Reflects the characteristics of the radiation field on the surface.
As the forecasting range expands to several hours or longer, due to atmospheric dynamics that change the cloud cover, simple steady assumptions are no longer valid, and the forecasting ability shifts from pure observation to the field of NWP models. The ability of the NWP model to predict the actual cloud shape fully illustrates the fidelity of the parameterization and the data assimilation of the dynamic and thermodynamic states used to initialize the model.
The NWP model should have the ability to represent the observed three-dimensional cloud field at the initial time, as well as include all necessary environmental conditions for maintaining and evolving cloud layers. However, this is still an unsolved problem so far. The core of this problem is the inherent flaws of NWP analysis. In a sense, the available observation data are not enough to describe the degree of freedom of the model. The information presented in the analysis is the result of a compromise between the model background and the actual environmental state. A cloud field may be similar (but not identical) to the observed value in terms of distribution, attributes, and evolution. The slight difference between the actual state and the simulated environment state will become larger and larger as the forecast time increases.
In terms of aerosol prediction, it is very important to combine accurate raw information (for example, biomass burning, dust storms, pollution) and active chemical reactions (different from passive advection of aerosols). At this time, satellites can provide critical information for mapping, monitoring, and eliminating the source of aerosols (for example, precipitation removal) on a global scale.
