ICACT20230158 Slide.15
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
|
ICACT20230158 Slide.14
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
If anyone has questions I’ll be happy to answer. Please send me an e-mail.
|
|
ICACT20230158 Slide.13
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
In conclusion, let me sum up my main points.
We developed PV power generation and power load prediction models using deep learning and machine learning.
And we rovided the validity of the results using eXplainable AI.
Prediction model was derived by substituting appropriate variables for the time when the power generation could not be measured.
Values influenced the result value was derived through the SHAP algorithm.
Research to increase energy efficiency in a small power grid by applying a prediction model developed in the future to a building or home unit and scheduling it based on the prediction results is expected to be possible.
|
|
ICACT20230158 Slide.12
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
This is the result of using TreeSHAP.
|
|
ICACT20230158 Slide.11
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
This is the result of the prediction model using LGBM. The low value of RMSE showed excellent performance.
|
|
ICACT20230158 Slide.10
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
These plots show how much each variable has affected the result value as a result of the DeepSHAP XAI algorithm.
In both models, we can see that solar power and consumption in the previous time had the greatest impact.
|
|
ICACT20230158 Slide.09
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
These figures are the results of two prediction models using LSTM.
Root Mean Square Deviation(RMSE) was used as an evaluation index, and the RMSE value was low, indicating excellent performance.
In the figure, the blue line represents the actual data and the orange line represents the predicted data, so you can see that most of the data matches.
|
|
ICACT20230158 Slide.08
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
The following figure is a workflow of a prediction model using LSTM and LGBM.
The other part is similar to the existing ai model, but the DeepSHAP and TreeSHAP shown in the figure are added to enable exploration.
|
|
ICACT20230158 Slide.07
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
This is an explanation of how we developed a predictive model.
First of all we create a new dataset by reflecting the time variables of meteorological data and power demand datasets to remove missing values and proceed with scaling of standardization and normalization of the data.
Second, we divide the dataset by a ratio of 80% and 20% to divide the training set for learning and the test set for testing .
Third, learning is carried out by putting the separated training set in the LSTM model.
|
|
ICACT20230158 Slide.06
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
LGBM is a gradient boosting framework that uses the decision tree algorithm among various machine learning algorithms.
The leaf-wise tree structure has the advantage of reducing losses compared to the level-wise tree structure and increasing accuracy while learning at high speed.
In this paper develops a PV power generation and power load amount forecasting model using LSTM and LGBM by reflecting an appropriate time variable.
|
|
ICACT20230158 Slide.05
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
The AI model has the following characteristics. One of the commonly used methods for time series forecasting is Recurrent Neural Networks(RNNs).
But, becomes a multiplication form with time and becomes gradient growth or exponential, learning to have dependency causes gradient vanishing or exploding problems, making training difficult.
To solve this problem, LSTM is used. In this paper, we used LSTM for Deep Learning
|
|
ICACT20230158 Slide.04
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
The table is the dataset we used in this paper.
There are a total of three data sets using weather data, PV Power Generation data, and Power load data. We add the discomfort index from the weather data.
The discomfort index was calculated using temperature and humidity.
We excluded summer and winter sunset times because the power generation data in the evening when the power generation cannot be measured makes the prediction worse.
|
|
ICACT20230158 Slide.03
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
Figure shows the basic behavior of xai.
In the picture on the left, possible to solve the problems of the existing ai model, find the cause factor, and understand how much it affected the results
In this paper, we use the SHAP model based on Shapley Value SHAP is based on Local Explanation, enabling global surrogates of the entire area of data.
|
|
ICACT20230158 Slide.02
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
This study started with the need for prediction of pv power generation and power load.
This is why we need pv generation and power load forecasting?
Global efforts, such as ESG or the Paris Climate Agreement, are underway and require a shift from traditional power to renewable energy.
PV of renewable energy is installed in existing buildings and maintenance is easy. However, PV is difficult and inefficient to provide reliable power due to weather and time. Power load forecasting is required for proper scheduling of PV and ESS
Number 2, Why we use xai.
As you are all aware Existing AI models have the problem called “BlackBox”, where results can be derived similarly to humans or as desired, but how and on what basis such results were produced are unknown. And Identify which variables have the most impact to address optimization issues. XAI can solve these problem.
|
|
ICACT20230158 Slide.01
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
The presentation will cover five contents.
I hope you will understand why I used xai when the presentation is over.
|
|
ICACT20230158 Slide.00
[Big slide for presentation]
[YouTube] |
Chrome Text-to-Speach Click!! |
 |
Hello, My name's Soohyun Shin and i'm going to talk about "Estimation of Power Generation and Consumption based on eXplainable Artificial Intelligence".
|
![[Big slide for presentation]](/upload/2023/0158/Presentation/15/15.jpg){kind=link}