Al Haromainy, Muhammad Muharrom (2021) Prediksi Data Time Series Multivariat Menggunakan Echo State Network Double Loop Reservoir Dan Harmony Search Optimization. Masters thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
05111850010044-Master_Thesis.pdf - Accepted Version Download (3MB) | Preview |
Abstract
Prediksi data deret waktu multivariat banyak diterapkan di berbagai bidang. Beberapa metode digunakan untuk penelitian prediksi data deret waktu, seperti metode Artificial Neural Network (ANN) dan Recurrent Neural Network (RNN). RNN menghasilkan nilai akurasi lebih baik dibandingkan ANN, karena dapat menyimpan memori (feedback loop) dengan melakukan perulangan dalam arsitekturnya kemudian menggunakannya untuk prediksi sehingga tidak membuang informasi dari masa lalu. Namun, salah satu kelemahan utama RNN adalah kesulitan dalam mengadaptasi bobot. Metode pengembangan dari RNN adalah Echo State Network (ESN) model prediksi deret waktu nonlinear multivariat yang kuat dan adaptif. Terlepas dari keunggulan yang disebutkan, pengaturan parameter ESN, seperti inisialisasi parameter reservoir harus dilakukan beberapa kali percobaan hingga mendapatkan hasil yang sesuai. Kemudian, struktur reservoir yang acak menyebabkan ESN membutuhkan waktu yang lama untuk menghasilkan reservoir.
Maka dari itu, penelitian ini mengusulkan Echo State Network dengan Double Loop Reservoir untuk proses prediksi data deret waktu multivariat dan dioptimasi menggunakan Harmony Search (HS), sehingga diharapkan dapat meningkatkan performa hasil prediksi. Metode evaluasi menggunakan Root Mean Square Error (MSE) dan Mean Absolute Percent Error (MAPE). Hasil prediksi menggunakan metode ESN-DLR lebih baik dari pada metode RNN dan ESN dengan nilai kesalahan 0.0001 dan 0.0082 untuk dataset 1, 5.88e-6 dan 0.0049 untuk dataset 2. Penentuan nilai parameter metode ESN yang sangat berpengaruh terhadap hasil prediksi, dibantu dengan optimasi metode HS sehingga mendapatkan nilai kesalahan lebih baik sebesar 3.36e-5 dan 0.0048 untuk dataset 1, pada dataset 2 memeroleh nilai kesalahan 1.46e-6 dan 0.0007.
=====================================================================================================
Multivariate time series data prediction is widely applied in various fields.
Several methods are used to research time series data predictions, such as the
Artificial Neural Network (ANN) and Recurrent Neural Network (RNN) methods.
RNN produces better accuracy than ANN, because it can save memory (feedback
loop) by looping in its architecture then using it for predictions so as not to throw
information from the past. One of the main drawbacks of the RNN, however, is the
difficulty in adapting weights. The development method of the RNN is the Echo
State Network (ESN), a multivariate and adaptive nonlinear time series prediction
model. Apart from the stated advantages, setting ESN parameters, such as reservoir
parameter initialization, must be carried out several times to get the right results.
Then, the random reservoir structure causes the ESN to take a long time to generate
a reservoir.
Therefore, this study proposes Echo State Network with Double Loop
Reservoir for the process of predicting multivariate time series data and is optimized
using Harmony Search (HS), so that it is expected to improve the performance of
the prediction results. The evaluation method uses the Root Mean Square Error
(MSE) and Mean Absolute Percent Error (MAPE). The prediction results using the
ESN-DLR method are better than the RNN and ESN methods with an error value
of 0.0001 and 0.0082 for dataset 1, 5.88e-6 and 0.0049 for dataset 2. Determination
of parameter values for the ESN method which greatly affects the prediction results,
is assisted by optimization HS method so that it gets better error values of 3.36e-5
and 0.0048 for dataset 1, for dataset 2 it gets error values 1.46e-6 and 0.0007.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | double loop reservoir, echo state network, reservoir computing, forecasting, harmony search double loop reservoir, echo state network, reservoir computing, forecasting, harmony search |
| Subjects: | H Social Sciences > HA Statistics > HA30.3 Time-series analysis Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > T Technology (General) > T57.84 Heuristic algorithms. |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis |
| Depositing User: | Muhammad Muharrom Al Haromainy |
| Date Deposited: | 01 Mar 2021 09:34 |
| Last Modified: | 04 Nov 2024 01:20 |
| URI: | http://repository.its.ac.id/id/eprint/82972 |
Actions (login required)
 |
View Item |