Бидний тухай
Багш ажилтан
When large-scale photovoltaic generation facilities are online, it becomes difficult to maintain the normal operation of the power grid system owing to the stochastic nature of solar irradiation. However, highly accurate solar forecasting lowers the uncertainty, thus supporting the reliable and economically beneficial functioning of the utility grid. The time horizon of 24 hours is of particular interest since the electric grid operation is planned with a focus on a day-ahead market. Therefore, we propose an hourly day-ahead solar irradiance prediction framework based on deep-learning, which has continuously shown promising results in time series forecasting. The global horizontal irradiance is measured for 3-5 years at 4 sites in the arid and cold climate of Mongolia, as shown in the figure. We preprocessed the collected data by filling in missing values with clear sky irradiance calculated by the pvlib library. For the data split, Ulaanbaatar and Erdenet sites are used for training, while Choir site is used for validation. Finally, Darkhan site is reserved for testing considering its similarity to Erdenet, characterized by a cold and dry winter. Since the model was exposed to this climate during the training, we can expect improved performance on test data. In short, the training, validation, and test data make up 58%, 23%, and 19% of the overall dataset. Once data preprocessing is completed, various deep-learning structures are benchmarked against the traditional statistical models in terms of mean absolute error (MAE) and root mean squared error (RMSE). Namely, naïve forecasting, linear regression, dense neural network (DNN), recurrent neural network (RNN), long-short term memory (LSTM), and convolutional neural network (CNN) are considered. On the one hand, the MAE and RMSE of the simple naïve forecast were 34.42 and 101.68 W/m2, respectively, which increased/decreased to 42.84 and 97.56 W/m2 for the multivariate regression. On the other hand, both error metrics are significantly reduced when deep learning models are implemented. Out of total 6 models, the best-performing model is found to be the sequence-to-sequence RNN model with 2 layers, 100 neurons, optimized by the stochastic gradient descent (SGD) under a learning rate of 1e-3. The resulting MAE and RMSE were 38.46 and 81.13 W/m2, respectively.
Accurate nowcasting of the global horizontal irradiance (GHI) would allow real-time monitoring of photovoltaic (PV) generation. Thus, it would have a positive impact on electricity dispatch by reducing the amount of optimal reserves. In this study, we aim to nowcast an hourly GHI considering its importance in the economically beneficial operation of the power utility. Figure 1 illustrates the proposed, deep-learning-based solar estimation framework. The multimodal input dataset is recorded at 4 ground sites for 3-5 years. It consists of various meteorological parameters and satellite data provided by the Japanese geostationary meteorological satellite Himawari 8/9, resulting in a total of 39 input features. In the data preprocessing routine, the original temporal resolution of 10 minutes is averaged into hourly values. Furthermore, the data is partitioned into the training, validation, and test sets, constituting 58%, 23%, and 19% of total data, respectively. In other words, we train the model with data from 2 ground sites. During the training, the model updates its internal weights as it learns the solar characteristics. Then, the model is validated at another site to check if it is actually learning or just memorizing the training data. Finally, the learned model’s generalization capability would be tested on independent data from the remaining site.
Performance Evaluation of the Semi-empirical Solar Estimation Model based on Satellite Data enhanced with Ground Measurement
Энэхүү судалгааны ажлаар оюутны танин мэдэх үйл явцыг тодорхойлох зорилгоор инженерчлэлийн чиглэлээр суралцаж буй 19 оюутныг санал асуулгад хамруулсан. Ингэхдээ 80 асуулт бүхий темпераментийг тодорхойлох асуулга, мэргэжлийн хичээлийн агуулга бүхий 20 асуулттай оюутны мэдээлэл боловсруулах хурдыг тодорхойлох ‘Kahoot’ тест болон 5 асуулт бүхий сургалтад ‘Kahoot’ тоглоомыг ашигласан сэтгэл ханамжийн судалгааг явуулсан. Судалгааны үр дүнгээр оюутны мэдээлэл боловсруулах хурд нь темпераменттэй хамааралтай буюу экстраверт оюутнуудын хариу үйлдэл үзүүлэх хурд нь интроверт төрлийн оюутнуудаас өндөр гэж гарсан. Мөн, мэдээлэл боловсруулах хурдыг тэмдэглэхэд ашигласан өрсөлдөөнт, асуулт хариултын ‘Kahoot’ тоглоомыг сургалтын үйл ажиллагаанд ашиглах нь эерэг үр дүнтэй гэж гарсан.
We aim to estimate solar resource of Mongolia using satellite data in combination with limited ground measurements. Visible channel images provided by Japanese Geostationary Meteorological Satellite (GMS) Himawari 8 are correlated with ground-based measurements of solar irradiation to derive parameters of the semi-physical model.
We aim to estimate solar resource of Mongolia using satellite data in combination with limited ground measurements. Visible channel images provided by Japanese Geostationary Meteorological Satellite (GMS) Himawari 8 are correlated with ground-based measurements of solar irradiation to derive parameters of the semi-physical model.
In this paper, we estimated diffuse fraction of global horizontal irradiation by implementing Erb’s model which computes diffuse fraction empirically from clearness index. At 4 locations across Mongolia, hourly, daily and monthly diffuse horizontal irradiation (DHI) values are estimated for year-long period and compared with the corresponding ground truth measurements. The results are demonstrated by different statistical parameters where normalized mean bias error (nMBE) and normalized root mean square error (nRMSE) were 2-20 and 10-33 percent, respectively. At all locations, the model underestimates diffuse irradiance which leads to negative bias and the correlation coefficients were greater than 0.87.
Many countries started to use renewable energy resource as an alternative energy source for electricity production because of environmental issues such as global warming and greenhouse gas concentration as a result of burning fossil fuels. Out of 5 main renewable energy resources, solar energy utilization is constantly increasing because of its mature technology and equally distributed resource throughout the world. Therefore, solar energy is used in many fields and in this study, photovoltaic system performance for meat freezer in rural areas of Mongolia is investigated. The main facilities are movable freezing container, grid connected 3kW photovoltaic modules, inverter, diesel generator and data logger to measure the parameters of this system. Starting from November 2015, we collected a year round site measurement which includes solar irradiation, outside and inside temperature of the container, electricity production and consumption. We use HOMER software to calculate the energy production by photovoltaic system and energy consumption of freezer system. Then, we compared this calculation with the real value and calculated the solar energy share of total electricity consumption. Also, we modeled an E-nose to monitor the freshness of meat preserved in the freezer. The result suggests that it is suitable to use photovoltaic system for meat storage in rural areas where grid electricity interrupts often.
