Fig 4 Optimization of hidden unit number
This paper sets the learning rate of the ELMAN neural network to 0.01, the training times to 1000, the target minimum error to 0.00001, the momentum factor to 0.01, and the minimum performance gradient to 0.00001.
Given the training duration of the algorithm model, this paper specifies the number of evolutionary iterations at 100 generations, the population size at 40, the maximum crossover rate at 0.85, the minimum crossover rate at 0.55, the minimum mutation rate at 0.35, and the number of individuals retained through the elitist strategy at 20.
To confirm the effectiveness of the proposed NGA-EMANCNN neural network prediction model, this study utilizes various statistical error measurement methods to evaluate the performance of the prediction model, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE).
NGA-LMANNCNN Photovoltaic Output Prediction Model Results: The paper divides the dataset into training and testing sets at an 8:2 ratio, using the training set to develop the prediction model and the testing set to assess the model’s performance and practicality subsequent to training. Given the runtime of the program, only the third ELMAN sub-neural network is optimized using NGA.
Taking January as an example, the output results of Network 1 on the test set are shown in Figure 5, the error figure is shown in Figure 6, and the calculation results of the model measurement index are shown in Table 1.