Existing Physics-Informed Neural Networks have drawbacks

The gap

Evidence profile

Sourced from the limitations and stated research gap of the source papers, classified as general, spanning 3 journals. Those papers have been cited 2 times in total.

Supporting evidence — 3 representative gaps

• Accurate State of Charge Estimation for Electric Vehicle Lithium-ion Batteries Using Shapley Additive Explanations and Kolmogorov Arnold Network (2026) · International Journal of Intelligent Engineering and Systems · doi

  low-temperature conditions, which led to inconsistencies in the predictions. tuning. However, extremely effectively requiring improved under they any Jafari [17] demonstrated three deep learning methods, Convolutional LSTM (ConvLSTM), LSTM and CNN with Particle Swarm Optimization (PSO), for improving the performance of SoC estimation in batteries. Initially, a CNN was utilized to extract spatial features from the battery data. Furthermore, LSTM was employed to capture the temporal dependencies between charge and discharge cycles. Finally, PSO was incorporated to fine-tune the hyperparameters and optimize model accuracy. The introduced model effectively predicted the SoC under varying environmental conditions. However, transient errors in SoC calculations caused sudden oscillations, which led to inconsistencies in the predictions under driving conditions. Mustaffa [18] established teaching-learning- based optimization with a deep neural network (TLBO-DNN) to enhance the accuracy of SoC estimation. Initially, 1,064,000 samples related to battery behavior were collected from the large datasets. Subsequently, a DNN model was designed to capture the complex nonlinear relationships in the collected battery data. Furthermore, the TLBO algorithm was introduced to optimize the biases and weights of the DNN. The presented TLBO-DNN accurately estimated the SoC of the EV; however, this approach was sensitive to the quality and representativeness of the training data, which tends to limit generalizability. Shahriar [19] presented a convolutional neural network with gated recurrent units and long short- term memory (CNN-GRU-LSTM) for estimating the SoC of an EV. Initially, battery data were collected from LG 18650HG2 lithium-ion battery cells, which were tested under different dynamic temperatures. Subsequently, the proposed deep learning models were employed to predict the EV SoC with low error.

  generallimitationsevidence 5/5
  Keywords: battery lstm conditions deep learning initially model tlbo collected inconsistencies predictions effectively convolutional optimization estimation
• An Adaptive-Weight Physics-Informed Neural Network Optimized by Grey Wolf Optimizer for Lithium-Ion Battery State of Health Estimation (2026) · Batteries · cited 2× · doi

  Existing Physics-Informed Neural Networks have drawbacks like single-feature physical constraints and rigid fixed-weight fusion. Insufficient time-series degradation modeling in existing methods. The need for a method that integrates battery degradation physics with deep learning, enabling strong fitting capability and physical interpretability.

  generalstated research gapevidence 5/5
  Keywords: existing physics-informed neural networks have drawbacks like single-feature
• NBE-VLT-PFO: hybrid deep learning transformer architecture for joint estimation of lithium-ion batteries (2026) · Frontiers in Artificial Intelligence · doi

  The need for accurate predictions of lithium-ion batteries' state of charge, state of health, and remaining useful life. The lack of effective methods for global hyperparameter tuning in deep learning systems for battery estimation. The requirement for a hybrid deep learning system that integrates multiple techniques for accurate battery estimation.

  generalstated research gapevidence 5/5
  Keywords: need accurate predictions lithium-ion batteries state charge health

Questions about this gap