Sale!

# CS 458 Final Project Solved

\$35.00

Category:

5/5 - (1 vote)

## Solar Power Generation Forecast

This project aims to develop solar power generation forecast models using the data to predict 24 h
ahead solar power generation on a rolling basis for three solar power plants located in a certain
region of Australia.

The data is from Global Energy Forecasting Competition 2014 (see
https://www.crowdanalytix.com/contests/global-energy-forecasting-competition-2014-
probabilistic-solar-power-forecasting). A portion of the datasets will be used in our project.

## 1. Dataset Description.

You will find the data in the excel file “solar.csv” in the webcampus. The first 5 data entries are
provided below:
The data includes weather forecasts for 12 weather variables, as obtained from the European
Centre for Medium-range Weather Forecasts (ECMWF). These variables are denoted as VAR
attributes (from VAR78 to VAR228) in the dataset, which are summarized in Table 1.

The hourly measurements for each solar power plant from 20120401 01:00 to 20140701 00:00.
are provided. ZONEID denotes the id of a solar power plant.

As we have three solar power plants,
ZONEID is from 1 to 3. The POWER attribute records the hourly solar power generation, which
has been normalized. For example, the first data entry gives the hourly measurements for solar
power plant 1 at 01:00 on April 1, 2012.

ZONEID TIMESTAMP VAR78 VAR79 VAR134 VAR157 VAR164 VAR165 VAR166 VAR167 VAR169 VAR175 VAR178 VAR228 POWER
1 20120401 01:00 0.001967 0.003609 94843.63 60.22191 0.244601 1.039334 -2.50304 294.4485 2577830 1202532 2861797 0 0.754103
1 20120401 02:00 0.005524 0.033575 94757.94 54.6786 0.457138 2.482865 -2.99333 295.6514 5356093 2446757 5949378 0 0.555

1 20120401 03:00 0.030113 0.132009 94732.81 61.29489 0.771429 3.339867 -1.98254 294.4546 7921788 3681336 8939176 0.001341 0.438397
1 20120401 04:00 0.057167 0.110645 94704.06 67.77528 0.965866 3.106102 -1.44605 293.2615 9860520 4921504 11331679 0.002501 0.145449
1 20120401 05:00 0.051027 0.18956 94675 70.17299 0.944669 2.601146 -1.90449 292.7329 11143097 6254380 13105558 0.003331 0.111987
Table 1. 12 weather variables

CS 458 Final Project

1. Split “solar.csv” into the training dataset “solar_training.csv” and the test dataset “solar_test.csv”
as follows:
• The training dataset is from 20120401 01:00 to 20130701 00:00.
• The test dataset is from 20130701 01:00 to 20140701 00:00.
2. Build a 24 h ahead solar power generation forecast model using the training dataset.
To give you an idea, an example for one solar power plant is given as follows. Let 𝑋𝑡 =
{𝑉𝐴𝑅78, 𝑉𝐴𝑅79, … , 𝑉𝐴𝑅228, 𝑃} denote a vector of all the attributes for one solar power plant.

Let 𝑓 denote the forecast model, which will predict 24 h ahead solar power generation by using
the current and historical measurements. For example, 𝑃̂
𝑡+24 = 𝑓(𝑋𝑡
), which uses only the current
measurements to predict 24 h ahead solar power generation, i.e., 𝑃̂
𝑡+24. You can explore the spatiotemporal correlation in the datasets to incorporate more relevant measurements as input for the
prediction. You can build a forecast model for each plant separately.

3. Model Evaluation.
You need to evaluate your model using the test dataset. Let 𝑃𝑡 and 𝑃̂
𝑡 denote the actual power
generation and the predicted power generation at time t, respectively. You will use the following
• Mean absolute error (MAE)
𝑀𝐴𝐸 =
1
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑖𝑛𝑡𝑠
∑|𝑃𝑡 − 𝑃̂
𝑡
|
𝑡

• Root mean squared error (RMSE)
𝑅𝑀𝑆𝐸 = √
1
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑖𝑛𝑡𝑠
∑|𝑃𝑡 − 𝑃̂
𝑡
|
2
𝑡

3. Report format.
• Cover page
o Title, You name, CS458 or CS658
• Introduction
o 1-2 paragraphs describing the “big picture” — similar to an “abstract”
o 2-3 paragraphs describing what exact problem you solve and goals to achieve

• Background and related work
o Review relevant papers, in which a brief (1-2 sentence) summary of each paper and ½-
2 pages summarizing what’s been done/tried before
o Hint: use scholar.google.com, and cite the papers properly.
o 1-2 pages describing your proposed solution to solve the defined problem and achieve
your goals, i.e., how you build your models (e.g., Data preprocessing, Model selection,
Parameter selection)

• Evaluation results
o Your test environment, tests/verification methods/steps
▪ Specifics on how to verify whether something work
▪ Specifics on how to test your approach

o Your experimental results: describe what experiments are done and provide
figures/tables and the corresponding explanations of the experimental results.
▪ Need to provide a table of MAE and RMSE and figures to compare the curves
of your predicted power with the actual power for days in different seasons.
▪ Discussions of your results (e.g., pros and cons for your prediction models)
Table Example

ZONEID 1 2 3 Overall
MAE
RMSE
• Conclusions