Difference between revisions of "Projects:2020s2-7522 Finding a relationship between predictability measure and actual prediction error using self-similarity measure"
m (→Methods) |
|||
| (5 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
| − | + | Predictability measure is an important aspect of judging the accuracy of prediction, and the value reflects the predicting precision. A small prediction error indicating a higher prediction accuracy, which is significant and useful in many fields. | |
| − | Predictability measure is an important aspect of judging the accuracy of prediction, and the value reflects the predicting precision. A small prediction error indicating a higher prediction accuracy. | + | Based on the existing PV power data, classifying method and analysing method, a relationship between predictability measure (not only restrict to Hurst Exponent) and prediction error will be discussed in this project. Based on the sorted data base, use MATLAB to perfome the mathematical operations and analysing them through time series regression model, exponential smoothing model and ARIMA model. Finally, analyse and compare the result given by the mathematical operations and forecasting models, then discussing the accuracy of power forecasting and pointing out the error analysis. |
| + | == Introduction == | ||
| + | The number of large-scale Photovoltaic (PV) and wind farms is rapidly growing in Australia and all around the world. There are many factors affecting the stability of renewable energy generation. Many researchers have developed forecast techniques to try to accurately predict the generation. However, it is still not accurate enough. This project will use predictability measures (such as Hurst exponent) and forecast models to predict the given data to evaluate the predictability of PV generation. Thus, find the relationship between predictability and actual forecast error. | ||
| + | |||
=== Project team === | === Project team === | ||
| Line 14: | Line 17: | ||
== Background == | == Background == | ||
| + | The number of large-scale Photovoltaic (PV) and wind farms is rapidly growing in Australia and all around the world. The majority of which came from small-scale rooftop PV. More than two million, or 21 percent, of Australian households now have rooftop solar PV. However, there are many factors affecting the stability of renewable energy generation. Many researchers have developed forecast techniques to try to accurately predict the generation. However, it is still not accurate enough. | ||
== Methods == | == Methods == | ||
| + | The first part is about forecasting. We will create three forecasting models based on the 80% of given data. Then, use the models to make a prediction of the remaining 20% data. We will calculate the prediction error (meaning absolute percentage error) by comparing the actual values with prediction values. | ||
| + | Hurst exponent will be used in this section. It is a measure of predictability based on self-similarity theory which is used as a measure of long-term memory of time series. The value of HE is between 0 and 1.When H approaches to 1, it means this data is more predictable. When H is smaller than 0.5, the data is not predictable which is random noise. We will calculate the 1000 sets of one year data in this project. | ||
| + | After calculating the forecasting error and HE. We will try to find the relationship between them. There are 1000 sets of data of PV generation from different areas in Australia. We will make a table and graph to see if there exists a relationship between prediction error and hurst exponent. Thus, evaluate this relationship and evaluate the predictability of PV generation. | ||
== Results == | == Results == | ||
Latest revision as of 22:42, 20 September 2020
Predictability measure is an important aspect of judging the accuracy of prediction, and the value reflects the predicting precision. A small prediction error indicating a higher prediction accuracy, which is significant and useful in many fields. Based on the existing PV power data, classifying method and analysing method, a relationship between predictability measure (not only restrict to Hurst Exponent) and prediction error will be discussed in this project. Based on the sorted data base, use MATLAB to perfome the mathematical operations and analysing them through time series regression model, exponential smoothing model and ARIMA model. Finally, analyse and compare the result given by the mathematical operations and forecasting models, then discussing the accuracy of power forecasting and pointing out the error analysis.
Contents
Introduction
The number of large-scale Photovoltaic (PV) and wind farms is rapidly growing in Australia and all around the world. There are many factors affecting the stability of renewable energy generation. Many researchers have developed forecast techniques to try to accurately predict the generation. However, it is still not accurate enough. This project will use predictability measures (such as Hurst exponent) and forecast models to predict the given data to evaluate the predictability of PV generation. Thus, find the relationship between predictability and actual forecast error.
Project team
Project students
- Erli Yin
- Xuan Hu
Supervisors
- Dr Ali Pourmousavi Kani
- Professor Mathias Baumert
Advisors
Objectives
Background
The number of large-scale Photovoltaic (PV) and wind farms is rapidly growing in Australia and all around the world. The majority of which came from small-scale rooftop PV. More than two million, or 21 percent, of Australian households now have rooftop solar PV. However, there are many factors affecting the stability of renewable energy generation. Many researchers have developed forecast techniques to try to accurately predict the generation. However, it is still not accurate enough.
Methods
The first part is about forecasting. We will create three forecasting models based on the 80% of given data. Then, use the models to make a prediction of the remaining 20% data. We will calculate the prediction error (meaning absolute percentage error) by comparing the actual values with prediction values. Hurst exponent will be used in this section. It is a measure of predictability based on self-similarity theory which is used as a measure of long-term memory of time series. The value of HE is between 0 and 1.When H approaches to 1, it means this data is more predictable. When H is smaller than 0.5, the data is not predictable which is random noise. We will calculate the 1000 sets of one year data in this project. After calculating the forecasting error and HE. We will try to find the relationship between them. There are 1000 sets of data of PV generation from different areas in Australia. We will make a table and graph to see if there exists a relationship between prediction error and hurst exponent. Thus, evaluate this relationship and evaluate the predictability of PV generation.
