基于深度神经网络的斜坡式重力储能系统质量块抓取装置控制方法

Control Method of Mass Block Grasping Device of Slope Gravity Energy Storage System Based on Deep Neural Network

重力储能系统对于质量块码放环节的时效性要求较高,针对现有抓取装置在时效性上的不足,提出了一种基于深度神经网络的斜坡式重力储能系统质量块抓取装置控制方法。首先,介绍了斜坡式重力储能系统的整体工作流程以及对质量块抓取装置的需求;其次,提出了根据质量块瞬时加速度计算其运行路程的方法,并根据质量块在缓冲平台的运行路程影响因素分析了该方法的误差来源,基于误差来源构造质量块运行路程的数据集;最后,引进深度神经网络,通过训练确定网络结构和参数,利用Dropout机制增强模型的泛化能力,得到质量块运行路程的预测模型。分析表明,所提控制方法可以较好地满足系统对码放环节的时效性以及精度的需求,使用本模型可在50 ms内给出质量块的运行路程预测值,且误差在±0.1 m内,验证了方案的可行性。

The gravity energy storage system has high timeliness requirements for the phase of mass block placement.Addressing the inadequacy of existing grasping devices in terms of timeliness,a control method for a slope gravity energy storage system's mass block grasping device based on deep neural networks is proposed.Firstly,the overall workflow of the slope gravity energy storage system and the requirements for the mass block grasping device are introduced.Secondly,a method for calculating the instantaneous acceleration of the mass block to determine its travel distance is proposed,and the error sources of this method are analyzed according factors affecting the travel distance of the mass block in the buffer platform.Based on these error sources,a dataset for the mass block's travel distance is constructed.Finally,a deep neural network is introduced,and the network structure and parameters are determined through training.The Dropout mechanism is employed to enhance the generalization ability of the model,resulting in a predictive model for the mass block's travel distance.The analysis indicates that the proposed control method can effectively meet the system's requirements for the timeliness and accuracy of the placement phase.The predicted value of the mass block's travel distance can be given within 50 ms,with a prediction error within 0.1 m,Which verifies the feasibility of the method.