Method for recognition of abnormal conditions using neural networks

A method of recognizing normal and abnormal conditions in real time of an object system and outputting the recognition, said object system having at least one neural network, said object system including at least one human operator providing input thereto, input data for said object system including input from said at least one human operator, said method comprising the steps of:

• setting bias and weight for each node of the neural network;
• inputting a plurality of sets of dynamic input trend data indicating normal operating conditions of said object system into said neural network, said normal operating conditions including operator errors;
• determining a plurality of sets of predicted output data based upon a sigmoid function of said plurality of sets of input data and said biases and weights;
• inputting a plurality of sets of output data indicating normal output results of said object system into said neural network;
• determining a difference between each set of predicted output data and each corresponding set of output data which indicates normal results;
• minimizing differences between input data and output data by changing the bias and weight for each node of the neural network;
• determining a deviation range for predicted output data based on the minimized differences;
• inputting past trend data into said neural network and into said object system;
• determining predicted output data from said neural network based on said bias and weight from each node;
• determining real-time actual output data from said object system;
• comparing differences between actual output data from said object system and predicted output data from said neural network;
• determining if the difference is within said deviation range indicating a normal real-time condition;
• determining if the difference is outside said deviation range indicating an abnormal real-time condition;
• and
• outputting a warning when said abnormal real-time condition exists in said object system based upon said determination of normal and abnormal real-time conditions.