基于特征空间变换与LSTM的中短期电煤价格预测

针对电煤价格影响因素多且非线性多时间滞后难以建模的问题，提出一种基于特征变换与LSTM的数据驱动的中短期电煤价格预测方法.为了充分挖掘海量数中蕴含的电煤价格规律，提出不同时间尺度颗粒度信息的特征变换方法;为解决多变量少样本造成过拟合，基于卡方分析和相关系数筛选中短期煤价的主要影响因素;以LSTM神经网络为基础，采用特征平移相关性分析方法确定不同影响特征序列的滞后性，通过主层次分析法优化模型中的信息冗余，在此基础上形成基于特征趋势的深度学习模型;利用多年历史数据及与多种模型的对比分析可知本文模型的有效性与准确性.

Medium and Short-Term Electricity Coal Price Forecast Based on Feature Space Transformation and LSTM

Aiming at the problem that there are many influencing factors of electricity coal prices and the nonlinearity and multiple time lags are difficult to model， a data-driven medium and short-term electricity coal price forecasting method based on feature transformation and LSTM was proposed. In order to fully excavate the electricity coal price law contained in the massive data， the feature transformation method of different time scale granularity information was proposed. In order to solve the problem of over-fitting caused by multiple variables and small samples， the main influencing factors of medium and short-term coal prices were screened based on chi-square analysis and correlation coefficient. Based on the LSTM neural network， the feature translation correlation analysis method was used to determine the hysteresis of different influencing feature sequences， and the information redundancy in the model was optimized by the main analytic method， and on this basis， a deep learning model based on feature trends was formed. The years of historical data and comparative analysis with various models show the effectiveness and accuracy of this model.