在全球能源绿色转型的背景下,页岩气作为低碳能源的重要性日益凸显,但其产量受高维、非线性及非平稳性因素影响,传统预测方法存在精度不足和计算复杂度高的问题。为此,本文提出一种基于RevIN-Autoformer-FECAM的深度学习模型,用于提升...在全球能源绿色转型的背景下,页岩气作为低碳能源的重要性日益凸显,但其产量受高维、非线性及非平稳性因素影响,传统预测方法存在精度不足和计算复杂度高的问题。为此,本文提出一种基于RevIN-Autoformer-FECAM的深度学习模型,用于提升页岩气产量预测的准确性。该模型通过可逆实例归一化(RevIN)缓解时间序列的非平稳性问题,结合Autoformer的自注意力机制捕捉长周期依赖关系,并引入频率增强通道注意力机制(FECAM)优化多频特征提取。实验基于威海页岩气田三口气井的生产数据,与Informer、Transformer等主流模型对比表明,RevIN-Autoformer-FECAM在均方误差(MSE)和平均绝对误差(MAE)指标上均显著优于基线模型,尤其在长周期预测(24~60天)中表现稳定。研究结果为复杂时序数据预测提供了高效解决方案,对页岩气开发优化具有重要应用价值。Against the backdrop of the global green energy transition, shale gas has emerged as a critical low-carbon energy resource. However, its production is influenced by high-dimensional, nonlinear, and non-stationary factors, while traditional prediction methods suffer from limited accuracy and high computational complexity. To address these challenges, this paper proposes a deep learning model, RevIN-Autoformer-FECAM, to enhance the accuracy of shale gas production forecasting. The model integrates Reversible Instance Normalization (RevIN) to mitigate non-stationarity in time series, leverages the self-attention mechanism of Autoformer to capture long-term dependencies, and introduces a Frequency Enhanced Channel Attention Mechanism (FECAM) to optimize multi-frequency feature extraction. Experiments conducted on production data from three shale gas wells in the Weihai field demonstrate that RevIN-Autoformer-FECAM significantly outperforms baseline models (e.g., Informer, Transformer) in terms of Mean Squared Error (MSE) and Mean Absolute Error (MAE), particularly showing stable performance in long-term predictions (24~60 days). The research provides an efficient solution for complex time series forecasting and holds significant application value for optimizing shale gas development.展开更多
苹果含有碳水化合物、维生素以及微量元素多种营养物质,有利于儿童的生长发育,增强记忆力,可以说营养价值在众多水果中堪称完美。在我国的苹果种植区内,陕西省是适合生产的最佳地区,并且随着时间的增长,陕西省苹果产量呈现出逐渐上升的...苹果含有碳水化合物、维生素以及微量元素多种营养物质,有利于儿童的生长发育,增强记忆力,可以说营养价值在众多水果中堪称完美。在我国的苹果种植区内,陕西省是适合生产的最佳地区,并且随着时间的增长,陕西省苹果产量呈现出逐渐上升的状态,其产量位居首位。以统计学、经济学为基础,采用ARIMA模型对陕西省苹果产量进行了分析预测研究。对陕西统计年鉴40年间陕西省苹果产量统计数据做统计分析,依据求和自回归移动平均模型理论,运用Eviews软件通过平稳检验、白噪声检验、模型识别、参数估计、模型的检验与优化等一系列的过程,最终建立ARIMA(2,1,1)模型。利用建立的模型对陕西省近3年的苹果产量进行了预测比较,对模型进行评估,其平均相对误差在6.61%,可以较好的反映陕西省苹果产量的发展趋势。使用ARIMA模型分析苹果产量等时间序列数据,可以利用其优势来预测未来的产量趋势,这对于农业生产规划和市场分析具有重要的实际应用价值。Apples contain carbohydrates, vitamins and trace elements that are beneficial for children’s growth and memory;in other words, their nutritional value is perfect among many fruits. In China’s apple planting area, Shaanxi Province is suitable for production of the best area, and with the growth of time, Shaanxi Province apple output shows a state of gradual rise. Its output ranks the first. Based on statistics and economics, ARIMA model was used to analyze and forecast the apple yield in Shaanxi Province. The statistical data of apple yield in Shaanxi Province in the past 40 years of Shaanxi Statistical Yearbook were analyzed. Based on the theory of autoregressive moving average model, the ARIMA(2,1,1) model was finally established by using Eviews through a series of processes, such as stationary test, white noise test, model recognition, parameter estimation, model test and optimization. The established model is used to forecast and compare the apple yield of Shaanxi Province in the past three years. The average relative error of the model is 6.61%, which can better reflect the development trend of apple yield in Shaanxi Province and make short-term forecast. On the basis of the above, some suggestions on improving apple yield in Shaanxi province are put forward.展开更多
文摘在全球能源绿色转型的背景下,页岩气作为低碳能源的重要性日益凸显,但其产量受高维、非线性及非平稳性因素影响,传统预测方法存在精度不足和计算复杂度高的问题。为此,本文提出一种基于RevIN-Autoformer-FECAM的深度学习模型,用于提升页岩气产量预测的准确性。该模型通过可逆实例归一化(RevIN)缓解时间序列的非平稳性问题,结合Autoformer的自注意力机制捕捉长周期依赖关系,并引入频率增强通道注意力机制(FECAM)优化多频特征提取。实验基于威海页岩气田三口气井的生产数据,与Informer、Transformer等主流模型对比表明,RevIN-Autoformer-FECAM在均方误差(MSE)和平均绝对误差(MAE)指标上均显著优于基线模型,尤其在长周期预测(24~60天)中表现稳定。研究结果为复杂时序数据预测提供了高效解决方案,对页岩气开发优化具有重要应用价值。Against the backdrop of the global green energy transition, shale gas has emerged as a critical low-carbon energy resource. However, its production is influenced by high-dimensional, nonlinear, and non-stationary factors, while traditional prediction methods suffer from limited accuracy and high computational complexity. To address these challenges, this paper proposes a deep learning model, RevIN-Autoformer-FECAM, to enhance the accuracy of shale gas production forecasting. The model integrates Reversible Instance Normalization (RevIN) to mitigate non-stationarity in time series, leverages the self-attention mechanism of Autoformer to capture long-term dependencies, and introduces a Frequency Enhanced Channel Attention Mechanism (FECAM) to optimize multi-frequency feature extraction. Experiments conducted on production data from three shale gas wells in the Weihai field demonstrate that RevIN-Autoformer-FECAM significantly outperforms baseline models (e.g., Informer, Transformer) in terms of Mean Squared Error (MSE) and Mean Absolute Error (MAE), particularly showing stable performance in long-term predictions (24~60 days). The research provides an efficient solution for complex time series forecasting and holds significant application value for optimizing shale gas development.
文摘苹果含有碳水化合物、维生素以及微量元素多种营养物质,有利于儿童的生长发育,增强记忆力,可以说营养价值在众多水果中堪称完美。在我国的苹果种植区内,陕西省是适合生产的最佳地区,并且随着时间的增长,陕西省苹果产量呈现出逐渐上升的状态,其产量位居首位。以统计学、经济学为基础,采用ARIMA模型对陕西省苹果产量进行了分析预测研究。对陕西统计年鉴40年间陕西省苹果产量统计数据做统计分析,依据求和自回归移动平均模型理论,运用Eviews软件通过平稳检验、白噪声检验、模型识别、参数估计、模型的检验与优化等一系列的过程,最终建立ARIMA(2,1,1)模型。利用建立的模型对陕西省近3年的苹果产量进行了预测比较,对模型进行评估,其平均相对误差在6.61%,可以较好的反映陕西省苹果产量的发展趋势。使用ARIMA模型分析苹果产量等时间序列数据,可以利用其优势来预测未来的产量趋势,这对于农业生产规划和市场分析具有重要的实际应用价值。Apples contain carbohydrates, vitamins and trace elements that are beneficial for children’s growth and memory;in other words, their nutritional value is perfect among many fruits. In China’s apple planting area, Shaanxi Province is suitable for production of the best area, and with the growth of time, Shaanxi Province apple output shows a state of gradual rise. Its output ranks the first. Based on statistics and economics, ARIMA model was used to analyze and forecast the apple yield in Shaanxi Province. The statistical data of apple yield in Shaanxi Province in the past 40 years of Shaanxi Statistical Yearbook were analyzed. Based on the theory of autoregressive moving average model, the ARIMA(2,1,1) model was finally established by using Eviews through a series of processes, such as stationary test, white noise test, model recognition, parameter estimation, model test and optimization. The established model is used to forecast and compare the apple yield of Shaanxi Province in the past three years. The average relative error of the model is 6.61%, which can better reflect the development trend of apple yield in Shaanxi Province and make short-term forecast. On the basis of the above, some suggestions on improving apple yield in Shaanxi province are put forward.
