不知道为啥用inverse_transform 反解析后数值会有一个量级的缩小：

Stock_LSTM.eval()

y_pre=Stock_LSTM(t_loader.dataset.x.cuda())
print(y_pre)

pre_y=y_pre.cpu().detach().numpy()[:,-1,0].reshape(-1,1)
print(pre_y)
pre_yy=scaler.inverse_transform(pre_y)

[L100/1000] Tra_Loss: 0.290563,Tra_acc: 0.537313||val_Loss: 0.040118,val_acc: 0.474576
[L200/1000] Tra_Loss: 0.203432,Tra_acc: 0.552239||val_Loss: 0.020833,val_acc: 0.474576
[L300/1000] Tra_Loss: 0.079258,Tra_acc: 0.567164||val_Loss: 0.008534,val_acc: 0.508475
[L400/1000] Tra_Loss: 0.025377,Tra_acc: 0.567164||val_Loss: 0.001823,val_acc: 0.576271
[L500/1000] Tra_Loss: 0.015414,Tra_acc: 0.582090||val_Loss: 0.001621,val_acc: 0.576271
[L600/1000] Tra_Loss: 0.012911,Tra_acc: 0.597015||val_Loss: 0.001221,val_acc: 0.610169
tensor([[[0.3775],
[0.3659],
[0.3631],
[0.3607]]], device=‘cuda:0’, grad_fn=)
[[0.36069888]]
[[758.95886]]
[[0.36314556]]
0.993262532905851

这里反向解析后获得的值为758.95886 与实际数值差距比较大
这是什么原因