Picked from the software, topic, and level of “Class LSTM Electricity Load Forecasting #6: MAPE/RMSE Evaluation and Visualization of Forecast vs Actual”.
At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
This is the final result, the LSTM algorithm from genetic algorithm predicts the load for 24 hours ahead throughout the year 2019 with an error of about 2.8 percent, much better than all other baselines. We also found the hardest day to predict, and the answer might be unexpected. Let's start. Hello, welcome to the final episode of the electricity load forecasting series with LSTM. The 2019 test data has never been touched during training or hyperparameter search, so the results reflect the actual performance. Today we fairly evaluate all models and analyze where the model is still wrong.
Step 1: Ensemble three seeds
Create evaluation.py file. A single training can be slightly lucky or unlucky due to random initialization, so each configuration is trained 3 times with different SID, and then the forecasts are averaged. This technique is called ensembling, and it usually also reduces the error. We compare the initial configuration from episode 4 with the configuration from the genetic algorithm, plus 3 baselines from episode 3.
Step 2: MAPE/RMSE Table & Hardest Day
Print the table, then find the hardest and easiest day. Run, 6 trainings take about 1.5 minutes. The results, naive about 4.4%, SARIMA 5.6%, REACH 3.3%, and both LSTMs in the range of 2.8%. The genetic algorithm found a configuration that is equal or slightly better than our manual choices automatically, and compared to REACH, the LSTM error drops about 1.5. The hardest day is usually a national holiday, for example, weapon steel stun on November 11th or Good Friday Monday, with an error of dozens of percent, because holidays rarely appear in the training data. Let's look at the graphs.
Step 3: Forecast Visualization & Error
Top left, LSTMGA and REACH forecasts for one week in May. Both follow the pattern, but LSTM is more accurate at the peaks. Top right, comparison of MAPE for all methods. Bottom left, hourly forecast error, lowest in the early hours and highest in the afternoon. Bottom right, monthly forecast error, highest in April and May which are full of holidays.
Done!
Done, and the series is also done. We have built a complete research on electricity load forecasting, data exploration, feature engineering and windowing, naive baseline, REACH, and SARIMA, LSTM with PyTorch, hyperparameter optimization with genetic algorithm, up to ensembling evaluation. For further research, add weather forecasting as input, special treatment for holidays, transformer models, or probabilistic forecasting. Thank you for following this series. Are you currently working on a thesis, dissertation, or research in the field of electrical load forecasting and time series? Cerdas Inovasi Digital helps implement deep learning, from data exploration, feature engineering, ARIMA and regression, LSTM with PyTorch, up to hyperparameter optimization with genetic algorithms, as well as offering deep learning courses with PyTorch. Want to master the entire series, from hourly load data to forecast evaluation? All six episodes, complete with step-by-step PDF guides and source code, are available to premium tutorial subscribers. Subscribe starting at Rp69.000 per month, save 28% with the 12-month package, or pay once Rp999.000 for lifetime access. Want just one class? Buy individual classes with lifetime access. Subscribe now at the premium page cerdasinovasidigital.com Contact us via WhatsApp, or visit cerdasinovasidigital.com Thank you for watching, see you in the next episode.
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,286 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
