In view of the fact that the demand for agricultural machinery equipment is affected by actual agricultural production and many other factors, and the demand data is cyclical and non-linear, making it difficult to accurately forecast the demand for agricultural machinery, a material demand forecasting method integrating SARIMA-improved RS-multistep LSTM was proposed. The seasonal differential autoregressive moving average (SARIMA) model was constructed by determining the parameter combination. The complete set Empirical Mode decomposition (CEEMDAN), improved random search (RS) algorithm and multistep short term memory network (LSTM) were introduced to construct an improved RS-Multistep LSTM model. The optimal weighted combination of SARIMA model and improved RS multistep LSTM model was used to obtain a combined prediction model. Using a certain model of agricultural machinery equipment as an example for verification. The results have showed that the proposed method can effectively predict the time series of the demand for agricultural machinery equipment, the evaluation indicators MSE, MAE and R² are 225.45, 13.22 and 0.920 9 respectively.

针对农机装备需求受实际农业生产和其他多种因素影响，需求数据呈现周期性、非线性等特点，难以准确预测农机需求的问题，提出一种集成SARIMA-改进RS-多步LSTM的农机装备需求预测方法。通过确定参数组合，构建季节性差分自回归滑动平均(SARIMA)模型。引入完全集合经验模态分解(CEEMDAN)、改进随机搜索(RS)算法和多步长短期记忆网络(LSTM)，构建改进RS-多步LSTM模型。将SARIMA模型和改进RS-多步LSTM模型进行最优加权组合，得到组合预测模型。以某型号农机装备进行实例验证，结果表明，所提方法能够有效预测农机装备需求的时间序列，评价指标MSE、MAE、R²分别为225.45、13.22和0.920 9。