The penetration of variable renewable energy generation is bringing about the new issues of variability to a power system operation. The issues are expected to be resolved through the optimum deployment of all the flexibility resources. The paper discusses about the variability and the flexibility to find the needs for the evolution of the forecast technologies and the system technologies of centralized and decentralized power systems.
This study presents a method of estimating the past total Photovoltaic Systems (PV) output results over a wide area, such as the supply area of an electric power company, and its associated precision. When estimating the total PV output over a wide area, it is difficult to grasp accurate information (PV direction, tilt, latitude, longitude, model of module and PCS, and solar irradiance incident on solar panel) for an estimate of the individual PV output. However, by using the mean as a substitute for the individual information, this method can increase the smoothing effect of the output errors and lower the total error.
In this paper, the authors propose a method to estimate total power output of photovoltaic (PV) systems connected in the distribution network. Total power output of PV systems connected to the distribution line is estimated based on the observed solar irradiation, statistical model and power flow in the distribution line, where PV systems are connected. The proposed method is divided into low and high-frequency component estimations, which estimate output of PV systems down to second and minute resolution depending on the temporal resolution of power flow. Performance of the proposed method is evaluated based on simulations using real data from two Japanese demonstration projects. For the estimation of output of PV systems in the distribution line, simulation results show that the root mean square error (RMSE) of estimation varies between 4 and 14 percents (normalized by PV systems capacity) according to power flow temporal resolution, PV penetration level, distance between solar irradiation observation point and distribution line. Up-scaling model is also investigated. Theoretical evaluation results show that having one representative point in every 40km × 40km distribution network is the most suitable configuration for estimating PV systems output for overall power system.
This paper describes an application of a neural network that is method forecasting to time variation of insulation intensity. In recent years, research and technological developments in the field of electrical energy have focused on photovoltaic. Therefore, the photovoltaic power generator is introduced in large quantities in the power system in the future is expected. However, despite the high expectations for renewable power generation technologies, it remains difficult to obtain stable power from such distributed sources, primarily because they depend on weather conditions and other variable factors. In order to apply to the supply and demand stable operation, we report a case of developing a method for predicting solar power generation using the gray theory and a neural network.
In this paper, forecasts of the daily global horizontal irradiance (GHI) obtained from the global spectral model (GSM) developed at the Japan Meteorological Agency (JMA) are evaluated using the surface-observed data at the six JMA stations for the four years during 2009-2012. Seasonal and regional characteristics of forecast errors of GSM GHI and the relationship between forecast errors and the length of the lead time of the forecasts are also investigated. The mean bias errors (MBE) of the monthly averaged GHI forecasts at Tsukuba station ranges from 0.5kWh/m2 to 2.1kWh/m2 per day and the root mean square errors (RMSE) of the forecasts ranges from 0.5 to 2.2kWh/m2 per day. Annual changes of both the MBE and the RMSE values are not large during the four years. Regional and seasonal characteristics of forecast errors are also found; negative biases in summer (positive biases in winter) are found in the Ishigakijima station, while positive biases are significant in all seasons at the other five stations. We also confirmed that RMSE values of the monthly GSM GHI are depended on the length of the lead time of the forecasts from 1-day ahead to 3-days ahead.
Because of the unstable power output characteristics of a photovoltaic power generation system (PVS), the on-line monitoring or nowcasting of the aggregated PVS power output is important for the stable operation of electric power system. In this paper, we propose a nowcasting method of the spatial average irradiance in the area of several km radius using the all-sky image and the observed irradiance both at single point. First, the proposed method classifies the sky condition based on the color information of the all-sky image. If the sky is classified as the uniform condition (very fine or very cloudy), the spatial average irradiance is estimated as the same with the single point observed irradiance. If not, the spatial average irradiance is estimated by using a neural network (NN) model utilizing the color information as the inputs. As a result of a demonstrative study for several months, the Root Mean Square Error (RMSE) of the NN model-based estimation for the periods of the non-uniform sky condition is 103.8W/m2 (15.2%). Because the performance of the sky condition classification is high, RMSE for the periods of the uniform sky condition is 41.8W/m2 (8.7%), resulting in RMSE of 77.4W/m2 (13.5%) for the entire periods.
In order to validate global horizontal irradiance (GHI) of the weather forecast from the Japan Meteorological Agency Meso-Scale Model (MSM), error causes of relatively large error cases of the forecast GHI in MSM is investigated around Tsukuba, Japan. The extracted 89 cases are categorized as overestimation or underestimation and are investigated in connection with weather conditions. For the case of overestimation, when the thick stratocumulus appears at the low level, the MSM cloud amount forecast fails to simulate such low-level cloud. This is the one of the reason why the forecast GHI becomes overestimation. For the case of underestimation, when the thin clouds appear through the all levels, the MSM fails to simulate cloud amount at middle- and high levels. This is the one of the reason why the forecast GHI becomes underestimated. In order to investigate the relationship between the irradiance forecast and the cloud amount forecast in details, sensitivity experiments of irradiance-related schemes are executed. Two types of experiments are made. One is the sensitivity experiment of the horizontal and vertical resolution in order to change the cloud distribution, which has the strong affection to the forecast GHI. The other is the sensitivity experiment of the effective radius of cloud water and ice affecting the interaction between the cloud and the solar radiation. Since both experiments have strong sensitivity to the forecasted GHI, it is important to develop the scheme which forecasts the cloud distribution and to choose the suitable effective radius of cloud water and ice.
Although renewable energies such as solar and wind power are penetrating rapidly, they require reserve powers for sudden change of output and sometimes cause serious incident in electricity grid because of their unpredictability. On the other hand home energy management systems also need efficient operation of energy storage devices to maximize the utility of facilities including photovoltaics. Therefore to know the uncertainty of their output is crucially important to utilize such unstable power sources effectively. As the solar irradiation has upper and lower bound, beta distribution which has finite interval of [0,1] is suitable to represent the uncertainty of the irradiation. Thus we applied beta regression, which is a kind of generalized linear models using beta distribution, to probabilistic forecast of solar irradiation. As a result of assessment of estimated models by using threat score, precision, and recall, the model employing variable dispersion beta regression showed the best performance in terms of predicting occurrences of large forecast errors. Moreover the precision parameter ϕ(x) of the variable dispersion beta regression was found to be an excellent indicator to alert possibility of large forecast errors.
The objective of this study is to analyze and to characterize the regional forecast error of hourly PV power generation in Japan. To make the analysis, data of 517 PV systems installed in 5 regions in Honshu Island were used to make 1 year of hourly forecasts. The regional forecasts of PV power were obtained from individual forecasts of the systems regarded in the study. Such forecasts were done using support vector regression and weather related information. The results showed that typical RMSE normalized by the region installed capacity were between 0.059 and 0.069kWh/kWcap. Values for the mean absolute error were near 60% of the RMSE being between 0.036 and 0.043kWh/kWcap. Furthermore, the smoothing effect was also characterized and values of 30% to 45% of error reduction were obtained according to the region and its PV systems characteristics. The study also provided values for the seasonal regional error in the Honshu Island, and it constitutes an important effort in the characterization of the regional forecast error of PV power generation in Japan.
With increased attention of renewable energy, large-scale installation of photovoltaic (PV) generation and electricity storage is expected to be installed into the power system in Japan. In this situation, we need to keep supply-demand balance by systematically using traditional power generation systems as well as the PV generation and storage equipment. Towards this balancing, a number of prediction methods for PV generation and demand have been developed in literature. However, such prediction-based balancing is not necessarily easy. This is because the prediction of PV generation and demand inevitably includes some uncertainty. Against this background, we formulate a problem to plan battery charge cycles while minimizing the fuel cost of generators with explicit consideration of prediction uncertainty. In this problem, given as parameter-dependent quadratic programming, the prediction uncertainty is described as a parameter in constraint condition. Furthermore, we propose a method to find a solution to this problem from the viewpoint of monotonicity analysis. Finally, by numerical analysis based on this problem and its solution method, we discuss the relation between the minimal reserve generating capacity and the required battery charge/discharge cycles to tolerate a given amount of prediction uncertainty.