Typical working conditions.

As a core component of the fully mechanized mining face, intelligent control of the shearer is fundamental to achieving unmanned mining and improving equipment reliability. To address the limitations of traditional optimization and deep reinforcement learning algorithms in achieving rapid and accurate self-adaptive control, this study proposes a novel shearer drum height control strategy based on the Deep Deterministic Policy Gradient (DDPG) algorithm. The 4602 workface at Yangcun Coal Mine and the MG2 × 55/250-BWD shearer model were used as engineering cases. A hybrid SVD-CWT and AlexNet transfer learning method was employed to identify coal and rock cutting states, achieving an accuracy of 95.06%. A DDPG-based self-adaptive hydraulic height adjustment model was then developed and validated through Matlab/Simulink and AMESim co-simulation, as well as a similarity-based physical test platform. Results show that the proposed method significantly outperforms conventional and fuzzy PID controls, reducing response time to 0.091 s and steady-state error to 0.00052 mm. Compared with TD3 and SAC algorithms, the system exhibited faster response, higher stability, and stronger anti-interference capability. The mean maximum error between simulation and experimental results was only 3.14%, confirming the feasibility and robustness of the proposed control strategy. This study provides a reliable approach for intelligent, adaptive height control of shearers under complex coal seam conditions.

作为综采工作面的核心组成部分，采煤机的智能控制是实现无人开采、提升设备可靠性的基础。针对传统优化算法与深度强化学习算法在实现快速精准自适应控制时存在的局限，本研究提出一种基于深度确定性策略梯度（Deep Deterministic Policy Gradient, DDPG）的新型采煤机滚筒高度控制策略。本研究以杨村煤矿4602工作面与MG2×55/250-BWD型采煤机为工程案例，采用混合奇异值分解-连续小波变换（SVD-CWT）与阿列克斯网络（AlexNet）迁移学习方法识别煤岩截割状态，识别准确率达95.06%。随后构建了基于DDPG的自适应液压调高模型，并通过Matlab/Simulink与AMESim联合仿真、基于相似性原理的物理试验平台对模型进行验证。结果表明，所提方法的性能显著优于传统控制与模糊PID控制，将响应时间缩短至0.091秒，稳态误差降至0.00052毫米。与双延迟深度确定性策略梯度（Twin Delayed Deep Deterministic Policy Gradient, TD3）和软演员评论家（Soft Actor-Critic, SAC）算法相比，本系统响应更快、稳定性更高、抗干扰能力更强。仿真与试验结果的平均最大误差仅为3.14%，验证了所提控制策略的可行性与鲁棒性。本研究为复杂煤层条件下采煤机的智能自适应调高控制提供了可靠的技术途径。