RNN and GNN based prediction of agricultural prices with multivariate time series and its short-term fluctuations smoothing effect

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Апрель 21, 2025

In this study, we investigate appropriate machine learning methods for predicting agricultural commodity prices. Since environmental factors including weather affect price fluctuations of commodities, constructed a multivariate time series dataset combining wholesale prices four commodities in South Korea, six variables, and week numbers. We adopted two prominent prediction based on recurrent neural networks (RNN) graph (GNN): one is the stacked long short-term memory, other consists GNN-based methods, spectral temporal network (StemGNN) convolutional network. Also, utilized univariate model as control to evaluate effectiveness approach investigation, applied five different smoothing window lengths effect mitigating predictive performance models. The experimental results showed that mitigation had greater impact improving models compared model. Among models, outperformed RNN-based view trained model, analyzed main variables affecting by utilizing adjacency weight matrices self-attention mechanism StemGNN.

Язык: Английский

---

A Hybrid Model Integrating Variational Mode Decomposition and Intelligent Optimization for Vegetable Price Prediction

Agriculture, Год журнала: 2025, Номер 15(9), С. 919 - 919

Опубликована: Апрель 23, 2025

In recent years, China’s vegetable market has faced frequent and drastic price fluctuations due to factors such as supply–demand relationships climate change, which significantly affect government bodies, farmers, consumers, other participants in the industry supply chain. Traditional forecasting methods demonstrate evident limitations capturing nonlinear characteristics complex volatility patterns of series, underscoring necessity developing high-precision prediction models. This study proposes a hybrid model integrating variational mode decomposition (VMD), Fruit Fly Optimization Algorithm (FOA), gated recurrent unit (GRU). The employs VMD for multi-scale original series utilizes FOA adaptive optimization GRU’s critical parameters, effectively addressing challenges high nonlinearity agricultural forecasting. Empirical analysis conducted on daily data six major vegetables, specifically, Chinese cabbage, cucumber, beans, tomato, chili, radish, from 2014 2024 reveals that proposed outperforms traditional methods, single deep learning models, models predictive performance. Experimental results indicate substantial improvements key metrics including Mean Absolute Error (MAE), Root Square (RMSE), Coefficient Determination (R2), with R2 values consistently exceeding 99.4% achieving over 5% enhancement compared baseline GRU model. research establishes novel methodological framework analyzing while providing reliable technical support monitoring policy regulation.

Язык: Английский

---

Forecasting Flower Prices by Long Short-Term Memory Model with Optuna

Electronics, Год журнала: 2024, Номер 13(18), С. 3646 - 3646

Опубликована: Сен. 13, 2024

The oriental lily ‘Casa Blanca’ is one of the most popular and high-value flowers. period for keeping these flowers refrigerated limited. Therefore, forecasting prices lilies crucial determining optimal planting time and, consequently, profits earned by flower growers. Traditionally, prediction has primarily relied on experience domain knowledge farmers, lacking systematic analysis. This study aims to predict daily at wholesale markets in Taiwan using many-to-many Long Short-Term Memory (MMLSTM) models. determination hyperparameters MMLSTM models significantly influences their performance. employs Optuna, a hyperparameter optimization technique specifically designed machine learning models, select Various modeling datasets windows are used evaluate performance with Optuna (MMLSTMOPT) predicting prices. Numerical results indicate that developed MMLSTMOPT model achieves highly satisfactory accuracy an average mean absolute percentage error value 12.7%. Thus, feasible promising alternative

Язык: Английский

---

Hybrid Deep Learning Model for Vegetable Price Forecasting Based on Principal Component Analysis and Attention Mechanism

Physica Scripta, Год журнала: 2024, Номер 99(12), С. 125017 - 125017

Опубликована: Окт. 18, 2024

Abstract With the aim of enhancing accuracy current models for forecasting vegetable prices and improving market structures, this study focuses on bell peppers at Nanhuanqiao Market in Suzhou. In paper, we propose a hybrid Convolutional Neural Network (CNN) Gated Recurrent Unit (GRU) model price based Principal Component Analysis (PCA) Attention Mechanism (ATT). Initially, utilized Pearson correlation coefficient to filter out factors impacting prices. Then, applied PCA reduce dimensionality, extracting key features. Next, captured local sequence patterns with CNN, while handling time-series features GRU. Finally, these outputs were integrated via ATT generate final prediction. Our results indicate that CNN-GRU model, enhanced by ATT, achieved Root Mean Square Error (RMSE) as low 0.1642. This performance is 11.11%, 15.79% better than PCA-CNN, PCA-GRU, CNN-GRU-ATT models, respectively. Furthermore, order prove effectiveness our proposed compared state-of-the-art classical machine learning algorithms under same dataset, deep shows best performance. Consequently, offers valuable reference

Язык: Английский

---