Trend Forecasting Method Based on Quantum Weighted Threshold Repeating Unit Neural Network

Abstract

The invention relates to a trend prediction method based on a quantum weighted threshold repeating unit neural network, comprising the following steps: constructing a threshold repeating unit model, a weighted neuron model with weight qubits and activity value qubits, and a quantum weighted threshold repeating unit neural network Network structure; real-time collection of original performance degradation vibration data; noise reduction processing of data by wavelet transform; extraction of permutation entropy information from denoised signals to form performance degradation index set; training and prediction of QWGRUNN network; forecast result. The invention introduces qubits to represent network weights and activity values, constructs a quantum phase shift gate weight matrix, and realizes the update of weight qubits and activity value qubits through the correction of gate parameters, which improves the network generalization ability, and further improves the The accuracy of forecasting the running trend of the monitored object; the use of dynamic learning parameters adapted to its own structure improves the convergence speed of the network and improves the calculation efficiency.

Description

technical field [0001] The invention relates to the technical field of information processing, in particular to a trend prediction method based on quantum weighted threshold repeating unit neural network. Background technique [0002] Rotating machinery (such as engines, steam turbines, etc.), as key equipment widely used in petrochemical, electric power, metallurgy, coal, nuclear energy, etc. Accidents that cause aircraft crashes and fatalities, resulting in heavy economic losses. Applying advanced fault diagnosis technology to the maintenance of rotating machinery can play an important role in ensuring the safe operation of equipment, saving maintenance costs and preventing environmental pollution, and has huge economic benefits. [0003] Generally, the maintenance methods of mechanical equipment can be roughly divided into three types: accident shutdown maintenance, regular shutdown maintenance, and condition-based maintenance (also known as predictive maintenance). Amo...

AI Technical Summary

Problems solved by technology

However, these prediction methods are difficult to obtain ideal prediction results due to their respective defects

For example, the AR model has poor fault tolerance; the kernel function and its parameters based on the least squares support vector machine are artificially selected in many cases, with many uncertainties; in the prediction method based on artificial neural network, such as BP neural network (Back- Propagation Neural Network, BPNN), Recurrent Neural Network (Recurrent Neural Network, RNN) and other classic neural networks, there are problems such as slow learning convergence speed, difficult training, and unstable learning and memory of the network.

The recently proposed Gated Recurrent Unit Neural Network (Gated Recurrent Unit Neural Network, GRUNN) overcomes the shortcomings of RNN gradient disappearance, but there are also problems such as difficult training process and insufficient generalization ability.

Images