I am looking some papers and share some experience about the use of grey neural network for irradiance forecast. The main point is the choice of the input and output variable to train the NN.
I have not seen experimental work undertaken in the forecasting of irradiance. Can you state the available parameters available?
In my research I want to find a methodology to set up an irradiance forecasting model based on neural networks in which uncertainty is directly taken into account. The approach is based on the use of an artificial neural network whose parameters are represented by grey numbers. The output of the proposed forecasting model is an interval (not a crisp value) which thus directly quantifies the imprecision/uncertainty or the vagueness of the forecasted values. I have long term ambient data (irradiante, temperature, wind speed) to train my neural network.
What is your objective? Is it to determine the irradiance for a given day of the year? As I understand, the data is not continuous for every day of the year. You intend to determine the irradiance for the day you do not have the recorded value and wish to estimate. It is proposed to model through ANN and for validation. In this process, certain uncertainty is bound to exist. Is my understanding correct? Let me know that first
The object is to forecast the irradiance value at certain time in the future starting from the knowledge of the previous measured values (every one minute) of irradiance and other ambient variables.
Specifically two types of forecast would be developed:
- Medium term (from about 20 to 10 hours a-head): the irradiance forecast value is hourly
- Short term (from about 4 to 1 hour a-head): the irradiance forecast value is a mean value over 10 or 15 min.
The forecast approach is based on grey numbers which allows for the characterisation of the uncertainty associated to the stage forecast produced by a neural network model. With this technique, the neural network parameters are grey numbers and thus the output of the neural network forecasting model at each time step is a grey number, i.e. an interval whose amplitude represents the total uncertainty (i.e. model, parameter and data uncertainty) related to the forecast.
- Badges
- Science topic