37results about How to "Guaranteed training accuracy" patented technology

The invention provides a parameter synchronization optimization method and system suitable for distributed machine learning. A machine learning algorithm achieved in a parameter server distribution mode is used for overcoming bottlenecks, such as a large amount of parallel machine learning training time delay caused by insufficient network and parameter server resources, of an existing algorithm in the parameter synchronization process. The system comprises a resource monitoring and distributing module at the parameter server end, a parameter maintaining module at the parameter server end, server resource request modules of all working nodes, parameter synchronization time interval control modules of the working nodes, non-synchronization time accumulation modules of the working nodes, parameter calculation modules of the working nodes and parameter synchronization modules of the working node. According to the parameter synchronization optimization method and system, different synchronization time intervals are selected for the different working nodes to avoid request emergency situations by monitoring resource occupancy conditions of a parameter server; meanwhile, it is guaranteed that the selected time intervals can meet the requirements for communication frequency reducing and training accurate rate guaranteeing at the same time, and the bottlenecks of an existing distributed machine learning system in the parameter synchronization process are effectively avoided.

The invention discloses a torque fluctuation compensation controller for a hybrid vehicle belt-type drive starter generator (hereinafter referred to as BSG) and a construction method thereof. The torque fluctuation compensation controller controls the compound controlled object, and the rotating The torque fluctuation compensation controller includes a PI regulator, an extended inverter control module and a current compensation module, and the composite controlled object includes a hybrid electric vehicle BSG, a photoelectric encoder disc, and a calculation module for rotational speed and electrical angle; the neural network is used to construct The current compensation module, the neural network adopts a four-layer structure, and uses the Powell algorithm to train the neural network. The method of the present invention can effectively suppress the current harmonics of the BSG motor by using the neural network to identify the current compensation module, significantly Reduce the torque ripple of the motor, so that the BSG system of the hybrid electric vehicle has good dynamic and static characteristics and strong robustness.

The invention discloses a method for predicting the performance of a centrifugal pump as a turbine based on an improved artificial neural network. The method firstly calculates the head of the optimal working point of the pump as a turbine at each specific speed in sections. H BEP,T ,flow Q BEP,T , calculate the flow rate of each operating point under the turbine state Q i and Q BEP,T The ratio a and Q i corresponding H i and H BEP,T the square root b of the ratio of Carry out L1 and L2 regularization at the same time; use the training set to train the artificial neural network; finally input the geometric parameters, a and b of the centrifugal pump to be predicted in the turbine state into the artificial neural network after training, and output Head and efficiency for each flow condition. The method of the invention has wide application range, high prediction accuracy and short calculation period.

The invention discloses a collaborative model training task configuration method for intelligent edge computing. The method is used for edge computing nodes and includes one or more training time slots, and each training time slot includes the following steps: moving to one or more The device sends a model training request; receives the available status and user data scale of the current time slot reported from one or more mobile devices; The number of training rounds required for interactive model training; interactive model training is carried out with the mobile devices participating in the training until the number of training rounds is reached. According to the training effect and the scale of user data reported by each mobile device, it is constructed to minimize the edge training resources The use-targeted optimization problem is solved and a new task configuration result is obtained. Compared with other methods, the training resource consumption of the present invention is much less, and the accuracy is not much different.

The invention discloses a third-party federated gradient boosting decision tree model training method. It includes the following steps: synchronous initialization of the training initiator and training participants; synchronous sampling of d sample data sets by the training initiator and training participants; binning of each feature data in their respective sample data sets by the training initiator and training participants , Record binning information and store bit slices; the training initiator calculates the first-order gradient sum and second-order gradient sum corresponding to each binning of each feature data of each sample data set, and the training initiator, training participants The party calculates the first-order gradient sum and the second-order gradient sum corresponding to each sub-bin of each feature data in the training participant's sample data set according to the safe multiplication protocol; the training initiator searches for the optimal split point, and synchronizes the results to the training participants side; repeat the above steps until the termination condition is met. The invention protects data security, reduces storage space, and greatly compresses communication traffic.