The data in normal working state is selected as the training set of the model, with a total of 12800 pieces. The data in a period of time before the fault alarm time is selected as the test set of the model, with a total of 5000 items. The input layer and output layer of the neural network are set as one neuron model, and the input layer and output layer are set as one neuron model. The model parameters, such as the number of hidden layer neurons, activation function, iteration times, the number of samples processed each time, dropout coefficient, optimization method, etc., are initially set. The mean absolute error (MAE) and root mean square error (RMSE) are used as the evaluation criteria of the model to optimize the model parameters to achieve the optimal prediction effect.
Where: n is the total number of sampling points; Y (I) is the actual temperature value of the ith sampling point; y ‘(I) is the model output value of the ith sampling point.
Some results of parameter optimization are selected. It can be seen that when other parameters remain unchanged, the effect of the model will be improved by increasing the number of neurons in the hidden layer within a certain range. With the increase of the number of iterations, the effect of the model will be improved to a certain extent, but the running time of the program will be longer. Finally, the number of hidden layer neurons is 65, the number of iterations is 550, and the number of samples is 60.