ST-GRU memristor neural network circuit for realizing gait prediction and training method

Abstract

The invention discloses an ST-GRU memristor neural network circuit for realizing gait prediction and a training method, and the circuit comprises a GRU unit circuit which is used for processing a time sequence and processing a time characteristic pulse voltage signal at a moment in the time sequence; acquiring an output result of the GRU unit circuit for processing the time sequence as a spatial characteristic pulse voltage signal, and inputting the spatial characteristic pulse voltage signal into the GRU unit circuit for processing the spatial sequence for processing the characteristic pulse voltage signal of the spatial sequence corresponding to the moment; an output result of the GRU unit circuit for processing the spatial sequence is input to the fully connected layer circuit. The problem that an existing GRU hardware circuit cannot train a target in multi-layer dynamic time sequence prediction is solved, multi-dimensional sequence prediction is achieved, the multi-dimensional sequence prediction is used for time sequence step length prediction, high accuracy and good calculation efficiency are achieved, and a large number of software computing power resources do not need to be consumed.

Description

technical field [0001] The invention relates to the field of neural network circuits, in particular to an ST-GRU memristive neural network circuit and a training method for realizing gait prediction. Background technique [0002] Much of the attention paid to artificial intelligence in recent years is due to advances in deep neural networks. There are many structures of deep neural networks, and recurrent neural network (RNN) is one of the important structures. There are many improved versions of the cyclic neural network, the more commonly used ones are long short-term memory (LSTM) and gate controllable recurrent unit (GRU), etc. These versions can effectively improve the gradient disappearance and gradient explosion problems in RNN; but due to the single The recurrent neural network has hysteresis, so it is not very effective when dealing with forecasting problems. [0003] Memristor is a nanoscale device with high volatility, high density, low power consumption, and co...

AI Technical Summary

Problems solved by technology

The present invention improves the deficiencies in the timing prediction of the existing GRU hardware circuit, solves the problem that the existing GRU hardware circuit cannot train the target in the multi-layer dynamic timing prediction, realizes the sequence prediction of multiple dimensions and It is used for the step size prediction problem of time series, has high accuracy and good calculation efficiency, and does not need to consume a lot of software computing power resources

Images