Short-Time Prediction of Solar Power Output Changes with Omnidirectional Camera

Introduction

Stable operation of photovoltaic generation requires real-time output forecasting. In solar power generation output prediction, satellite cloud observation data are used. This system enables large-scale solar power forecasting for mega solar. However, in recent years, with the spread of solar power generation, more accurate and localized energy management has become more important. From such background, FURUNO has developed a prototype for predicting change in local solar radiation using omnidirectional camera.

Case study

Images taken by the omnidirectional camera (Fig.1-a) are transferred to the server every minute. The transferred picture is analyzed in the receiving server to determine the cloud on the image, and a short-term change in the amount of solar power generation is predicted. A cloud measurement system was installed next to the solar panel installed at DAIHEN Corporation (Fig.1-b). The maximum output of solar panel is 82.6kW, and the installation area is 15 x 36 m. In this experiment, the amount of solar power generation and the prediction result of the cloud monitoring system were compared at the same point.

Results

Fig.3 shows the amount of power generation (PV output unite: kW) generated by the solar panel on July 24, 2018. The amount of power generation has repeatedly increasing and decreased rapidly from 10:00~11:00. At 15:00, there was a sharp drop in power generation, but only once. The comparison of the satellite image and the omnidirectional camera image at 15:00 is shown in Fig.4. The red * mark is the observation position of a cloud camera. In the satellite image, a small cloud of about few pixels is captured with high resolution by the cloud camera. The clouds that mask the sun are very small, but this causes a sharp drop in power generation Fig.4. Therefore, it is necessary to capture the detailed advection of clouds in order to predict rapid changes of power generation.

In this case, the cloud advection is estimated from the image at multiple times. Then, it predicts the rate at which clouds mask the sun at a given time (referred to as “sun forecast index” in this presentation). If this value is low, it indicates that there is a high possibility of clouding or rain, and if the value is large, it indicates that the possibility of sunny is high. Here, the solar prediction index is calculated using only past information for prediction, and solar power generation amount itself is not predicted. The prediction time is up to 10 minutes later.

Fig.5 (a) shows the prediction results of the cloud camera system. Fig.5 (a) shows the result of predicting the moving direction of the cloud from the image. The pixels determined to be clouds are colored, a yellow circle indicates the position of the sun, and green arrow indicates the movement direction and distance after one minute.

Fig.5-b shows the prediction result of time series. The vertical green line at 10:09 represents the predicted start time, the graph solid black line represents the past solar power generation output, and the dotted black line represents the actual solar power generation amount after the prediction.

The red solid line shows the solar power prediction index. Also, the lower part shows the images at each time (10:08, 10:11, 10:14, and 10:18). When the solar prediction index (red line) of Fig.5 (b) and the actual power generation output (black dotted line) are compared, it can be seen that the tendency is matched well. For example, it is predicted that it will be cloudy at 10:15 and clear at around 10:19 according to the solar prediction index. In fact, it can be seen that the power generation output is low at 10:15 and the power generation output is high at 10:19, and it can be predicted correctly. This result suggests that cloud cameras can be used to predict changes in solar radiation.

Conclusions

In this presentation, we report the prediction of solar power generation in a short time using a cloud camera system. As a result, it was shown that the tendency of the output change of the solar power generation after 10 minutes was predicted by analyzing the image of the omnidirectional camera.