The invention relates to a tower crane real-time transverse control device and method based on deep reinforcement learning, and the method comprises the following steps: obtaining the model of a tower crane, actual structure information, the mass of a balancing weight and the mass of a hoisted weight, and constructing a training virtual environment based on the obtained information; the method comprises the following steps: acquiring positions and motion information of a lifted weight and a balancing weight on a tower crane in a training environment, and building a deep reinforcement learning neural network model; constructing a mechanically-driven reward function and two-stage training of vertical lifting and axial movement; setting an early warning method of a moment limit value and an inclination angle limit value, and establishing risk assessment of the whole hoisting process of the tower crane; based on a trained model, the transverse control of the tower crane is intelligently decided, and based on monitoring information, the transverse movement of the balancing weight is automatically decided. According to the method, the stress state of the tower crane is intelligently optimized, the requirement for a foundation during design is reduced, the tower crane is in a reasonable state of automatic adjustment all the time, and therefore the stability and safety of construction are improved.
本发明是一种基于深度强化学习的塔吊实时横向控制装置及方法,方法包括以下步骤:获取塔吊型号、实际结构信息、配重块的质量、起吊重物的质量,基于获取的信息构建训练虚拟环境;获取训练环境中起吊重物、配重块在塔吊上的位置及运动信息,搭建深度强化学习神经网络模型;构建力学驱动的奖励函数以及垂直起吊与轴向运动的两阶段训练;设置力矩限值与倾斜角度限值的预警方法,建立塔吊起吊全过程的风险评估;基于训练的模型,智能决策塔吊的横向控制,基于监测信息,自动决策配重块的横向运动。本发明智能优化了塔吊的受力状态,减小了设计时对基础的要求,让塔吊时刻处于自动调节的合理状态,从而提高施工的稳定性和安全性。
Tower crane real-time transverse control device and method based on deep reinforcement learning
一种基于深度强化学习的塔吊实时横向控制装置及方法
03.09.2024
Patent
Elektronische Ressource
Chinesisch
Deep learning-based tower crane real-time path planning system and method
Europäisches Patentamt | 2023
|Positioning device for tower crane, tower crane control device and control method
Europäisches Patentamt | 2022
|