Collected NVIDIAGTX1060 GPU. Each algorithms were trained 100 instances underand exact same experimental
Collected NVIDIAGTX1060 GPU. Each algorithms have been trained one hundred times underand similar experimental conditions. sets of experiment, we usedthe The prediction final results the original load data from the five In our extrusion cycles in 25 sets of extrusion cycleFigure 8. Below the exact same experimental environment and education test set are shown in information collected in the 1# measuring point to make predictions. The prediction results andresults ofload load in the 5 sets of extrusion cycles inside the test set instances, the prediction original the data data for the duration of the service course of action from the extruder are shown in Resulting from 8. Beneath precisely the same experimental environmentgradient explosion, the could be noticed. Figure the troubles of gradient disappearance and and instruction occasions, the prediction outcomes from the loadcan not meet the prediction requirements inside the burst be seen. unmodified RNN algorithm data in the course of the service method on the extruder can stage of As a result of the complications hasgradient disappearance and gradientfalling trend. The predicted data, although there of been a slight fitting inside the rising and explosion, the unmodified RNNof LSTM algorithm has comparable extrusion needs in thewith the actual extrusion load algorithm can not meet the prediction cycle traits burst stage of data, althoughand the predicted final results are inside the rising and falling trend. The predicted load of load, there has been a slight fitting closer towards the actual data, which reflects the strong LSTM algorithm has equivalent extrusion cycle characteristics with the actual extrusion load, memory and understanding ability of LSTM network in time series. as well as the predicted outcomes are closer towards the actual information, which reflects the powerful memory and understanding capability of LSTM network in time series.Appl. Sci. 2021, 11, x FOR PEER Critique Sci. 2021, 11,eight of 13 8 ofFigure eight. Comparison of forecast results and original information. Figure eight. Comparison of forecast results and original data.In line with the prediction result Diversity Library Physicochemical Properties indicators the two models around the test set, the the According to the prediction result indicators ofof the two models on the test set, loss function values of distinctive models are shown in Table Table 1. The RMSE RMSE andvalues loss function values of different models are shown in 1. The MSE, MSE, and MAE MAE of LSTM LSTM and RNN algorithm are 0.405, 0.636, 0.502 and four.807, 2.193, 1.144, respecvalues of and RNN algorithm are 0.405, 0.636, 0.502 and 4.807, 2.193, 1.144, respectively. It is identified is located that compared with RNN model, the information error of LSTM network is tively. It that compared with RNN model, the predictionprediction information error of LSTM closer to is closer larger The greater prediction accuracy additional reflects the prediction network zero. Theto zero.prediction accuracy additional reflects the prediction performance of LSTM network, so LSTM model can far better adapt towards the predicament of random load prediction functionality of LSTM network, so LSTM model can improved adapt for the situation of ranand meet the demands of load SBP-3264 Protocol spectrum extrapolation. dom load prediction and meet the wants of load spectrum extrapolation.Table 1. Comparison of prediction performance amongst LSTM and RNN. Table 1. Comparison of prediction functionality between LSTM and RNN. Model Model RNN RNN LSTM LSTM MSE MSE four.807 four.807 0.405 0.405 RMSE RMSE 2.193 two.193 0.636 0.636 MAE MAE 1.144 1.144 0.502 0.four. Comparison of Load Spectrum four. Comparison of Load Spectrum 4.1. Classification of Load Spectrum 4.1. Classification of Load.