Auto-compensated electrostatic induction probes are commonly used to measure the electric potential at the surface of corona- or tribo-charged bodies. The aim of this paper is to propose a method to determine the actual surface potential distribution using stochastic search algorithms combined with neural networks. This technique leads to a multi-layer perceptron neural network (MLP) that approximates closely the actual distribution of the potential in any point on the surface. At First, the measured values are approximated by a radial basis function neural network (RBFNN) that generates the necessary data for the stochastic search algorithm and 2-dimensional convolution. Secondly, the convolution is performed using the point spread function (PSF) of the probe and iteratively the actual potential at every point on a uniform grid over the surface is determined. In fact, the stochastic search algorithm generates some candidates of the actual distribution of the potential and picks the best one by evaluating certain criteria. In the third place, the obtained distribution of the potential is de-noised and smoothed by training a multi-layer perceptron neural network (MLP) that has, as input, the position of a point on the surface and, as output, the value of the actual potential in that point. The proposed technique was tested and validated theoretically and experimentally. A comparison with the method based on the inverse of the PSF of the probe was performed. The obtained MLP can accurately predict the actual potential at any point on the surface investigated.

自动补偿静电感应探头通常用于测量电晕或摩擦带电体表面的电势。本文的目的是提出一种使用随机搜索算法结合神经网络来确定实际表面电位分布的方法。这种技术导致了一个多层感知器神经网络 (MLP)，它非常接近表面上任何点的实际电位分布。首先，测量值由径向基函数神经网络 (RBFNN) 近似，该网络为随机搜索算法和二维卷积生成必要的数据。第二，卷积是使用探针的点扩散函数 (PSF) 执行的，并迭代地确定表面上均匀网格上每个点的实际电位。事实上，随机搜索算法会生成一些潜在的实际分布的候选者，并通过评估某些标准来选择最佳的一个。第三，通过训练多层感知器神经网络 (MLP) 对获得的电位分布进行去噪和平滑处理，该网络具有作为输入的表面上点的位置，以及作为输出的值在那一点上的实际潜力。所提出的技术在理论上和实验上进行了测试和验证。与基于探针的 PSF 倒数的方法进行了比较。