提高采煤机记忆截割精度对于实现采煤机截割滚筒的自动调高，提升采煤机自动化水平具有重要意义。针对目前采煤机传统记忆截割方法精度不高的问题，根据采煤机截割过程具有一定重复性的特点，提出了一种基于深度长短时记忆(Long Short-Term Memory，LSTM)神经网络的采煤机记忆截割轨迹预测方法，通过MATLAB平台实现了模型的构建与模型参数的优化，并使用实际截割数据对深度LSTM神经网络模型进行了验证。预测实验的结果表明，深度LSTM神经网络相对于支持向量回归与梯度提升回归树算法在截割轨迹预测方面具有更高的准确性。深度LSTM神经网络的平均绝对误差、平均绝对百分误差、均方根误差均低于支持向量回归与梯度提升回归树算法。实际生产中采煤机需进行多次循环截割，考虑到实时性问题，神经网络模型需要对截割轨迹进行多步预测。为了进一步提升模型能力，提高模型在进行多步预测时的准确性，提出了一种LSTM神经网络的改进结构。通过在LSTM神经网络中引入比例因子，强化了神经网络的记忆保持能力。缓解了随预测步数增加，深度LSTM神经网络模型预测误差增大的问题。并对改进后模型与原模型进行了预测对比实验，实验结果表明，相较于未改进的模型，改进后的预测模型在多步预测中具有更好的表现，且优势随着预测步数的增加而更为明显。

It is of great significance to improve the precision of memory cutting of shearers,which is im-proving the automation of shearer.In view of the low precision of traditional shearer memory cutting and the repeatability of shearer cutting,a cutting path prediction method based on long short-term memory (LSTM) neural network was proposed.In MATLAB,the model was validated with real cutting data.The prediction results show that the deep LSTM has a higher accuracy than other machine learning algorithms,such as support vector regression and gradient boosted regression trees.The mean absolute error,mean absolute percentage error,and root mean square error of deep LSTM are lower than those of support vector regression and gradient boosted regression trees.In order to further improve the ability of the model in multi-step prediction,the scale factor was introduced to improve the structure of LSTM neural network.The scale factor can improve the ability of LSTM to keep memory and alleviating the problem that the error of deep LSTM neural network increases with the increase of prediction step.The experimental results show that the improved prediction model performs better in multi-step prediction.