Indicator diagram identification based on ART2 neural network and features of moment invariant

The indicator diagram is an important basis to analyze the downhole conditions of the pumping unit. Currently, it is mainly depends on artificial judgment to identify the indicator diagram, that not only needs higher requirements of the peoples´ experience, but also spends much more time and energy. ART2 neural network for an arbitrary sequence of continuous or binary mode, has the ability of fast and stable learning. It overcomes the shortcomings of the feed-forward neural networks, such as learning slowly; falling into the local minimum and washing away previously learned information easily. Moment invariants can reflect the key characteristics of two-dimensional graphics; it also owns the traits of rotation, stretching invariance and strong anti-interference, etc. It can be used for the feature extraction of pattern recognition. This paper mainly based on ART2 neural network and the moment invariant to identify the indicator diagrams in different working conditions. The recognition result shows that the ART2 neural network owns many advantages such as the fast, stable and high accuracy performance in the indicator diagram recognition process.