Radial Basis Function Neural Network

The RBFNN consists of an input vector succeeded by a layer of the radial basis function neurons and the resultant output layer with one node per group. Categorization is executed based on the measurement of the input data’s resemblance with the data points of the training dataset. Each of the neurons store a prototype which will be one of the samples from the training set.

When we classify a new n-dimensional input vector,the euclidean distance between the input and the prototype is computed for each neuron. For instance, if there are two categories (say category A and B) present in our training set then we calculate the closeness of our input data with both the given groups by means of their euclidean distance. If the distance measured between the input data and the category A is smaller as compared to the distance between the input data and category B, then the input data is classified as class A.

Every neuron of RBF compares the input data with the prototype of the samples in the training dataset. The resultant value is a measure of similarity between the range of 0 and 1. When the input data is similar to its prototype then the resultant value of the RBF neuron will be 1 and the euclidean distance between the two will exponentially dive to zero.

Radial Basis Neural Network

Applications of Radial Basis Function Neural Network

• Power Restoration: There is a possibility of an over-voltage power cut that may happen due to the transformation switching. The over-voltages may harm some equipment as well as cause a delay in the restorations of the power systems. A developed RBFNN model gives the equivalent attributes as the input to the model. It trains for the worst case situation that may occur like switching angles or the remainder flux. The model is then tested for such cases. It ensures the power-cut that happens to the supply that connects the most number of houses is restored first and eventually it goes on until the supply that impacts only one house is restored in the last.
• Mechanical Fault Diagnosis of a Gearbox: We have had a brief analysis over how the RBFNN is helpful for the pattern classification and contributes to a quick learning pace along with the capability to map in a non-linear fashion. These features help in employing the fault diagnosis. The gearbox is a very frequently used equipment in the field of engineering. By means of a trained RBFNN model, we can ensure that the diagnosis of the mechanical faults can take place correctly.