| 摘要 |
PURPOSE: A method for controlling cooling of wire rods by neural network is provided which performs cooling controlling on the wire rods in real-time according to condition change of a cooling facility by learning new operation patterns after receiving material information and operation process values from a cooling zone. CONSTITUTION: The method for controlling cooling of wire rods by neural network comprises steps of converting the received information into information for neural network after receiving material information and process information of a cooling facility as field information, classifying cooling patterns by statistically averaging the information after collecting the information in a certain intervals, and judging whether the cooling patterns are belong to the learned pattern domain or not(41¯45); storing and learning the connection strengths if an error between the output value and the target value is converged into the setting range by repeating a process of correcting connection strengths between each layers so as to reduce an error between an output value calculated and outputted into an output layer and a target value after inputting the information as an input layer of a cooling learning neural network consisting of connection strengths and critical values for transmitting an output value of the front layer node between input layer, concealed layer, output layer and each layers as an input of a back layer node in case of information of non-learning patterns in the judgement step(52¯55); and deducing new set values of the cooling facility from the output layer by inputting the field information as an input layer of a cooling setting neural network in which connection strengths and critical values between each layers of input layer, concealed layer and output layer are connection strengths and critical values that are stored in the information patterns in case of information of learning patterns in the judgement step, and deducing the set values by selecting approximate cooling patterns in the cooling pattern table in case of information of non-learning patterns in the judgement step(47¯51).
|
