I need to perform single-step prediction of voltage values using data from the past 6 time nodes.

I need to perform single-step prediction of voltage values using data from the past 6 time nodes. However, there are errors in the data input that are causing incorrect prediction results. How should this be handled?"​"Given that my PyTorch model has an input shape of [batch_size, 6, 4] and an output shape of [batch_size, 1], how should I preprocess the upstream data to ensure compatibility and successful model prediction?"

[Chuguang Pan]

@丁. According to the example presented in https://ww2.mathworks.cn/help/deeplearning/ug/predict-responses-using-onnx-model-predict-block.html, it is important to set up the Sample Time when simulate ONNX model in the simulink environment as a dynamic system. In my opinion, your ONNX prediction model has valid input data when the entire model has been simulated forwad 6 steps. Maybe you can debug the simulation model step by step using the Simulink DeBug functionality and inspect the input data of ONNX model to see how the input data change.

[Sam Chak]

@Chuguang Pan. There appears to be no error message, as Simulink successfully displays the output value at the Display block. However, the predicted result likely differs from the output produced by the PyTorch neural network model in Python.

[丁]

Yes, currently the PyTorch model achieves accurate training results. However, when deployed in Simulink for prediction, the output is incorrect and exhibits erroneous behavior. The sampling time is set to 0.01 seconds per step, and data dimension transformation is applied, yet the prediction results remain inaccurate.

[Sam Chak]

Hi @丁

If the data are confidential, create a toy problem (for example, a two‑input dataset) and train the PyTorch model. After verifying the PyTorch model achieves accurate training results, deploy it in Simulink. If the erroneous behavior persists, post both the Simulink file and the PyTorch model in ONNX format so that experienced neural‑network users can assist.

[丁]

This research component comprises Python scripts, ONNX model files, and associated Simulink simulation models. I shall accord utmost consideration to all recommendations offered and will duly document your valuable contributions in my subsequent research records.