Background: The Equivalent Salt Deposit Density (ESDD) is the basis of determining
pollution classes and mapping grid pollution areas. The influence of environmental factors on the ESDD
is complex, and it is difficult to establish an accurate mathematical model to fit the nonlinear
relationship between them.
Methods: In order to predict effectively the ESDD, a model of Adaptive Quantum Particle Swarm
Optimized BP neural network (AQPSO-BP) was proposed. In this algorithm, the encoding mechanism
based on probability amplitude of quantum bits was used to expand the ergodicity of population. The
position and velocity information of each particle was applied to adaptively adjust the inertia factor. At
the same time, the non-linear dynamic adjustment strategy of acceleration factors and mutation
operation were introduced to reduce the probabilities of trapping in the local optima solution.
Results: The prediction results show that the average relative error, the mean absolute error, the mean
squared error and the coefficient of determination are 0.1393%, 1.27E-04, 2.33E-06 and 0.9830,
respectively; the average relative variance is 0.0171.
Conclusion: Compared with the Particle Swarm Optimized BP network (PSO-BP) and Quantum
Particle Swarm Optimized BP network (QPSO-BP) models, the AQPSO-BP algorithm has higher
prediction accuracy and stronger generalization ability, and is suitable for evaluating the contamination
level to prevent flashover on polluted insulators.