Grip　force　prediction　plays　an　important　role　in　biomechanical　research,　sports　medicine,　and　clinical　rehabilitation.　Most　of　the　current　studies　in　this　area　only　focuses　on　the　characteristic　input　of　surface　Electromyography　(sEMG)　signals,　but　the　acquisition　and　processing　of　sEMG　are　complicated　and　vulnerable　to　electromagnetic　interference.　The　impedance　signal　has　the　advantages　of　easy　acquisition　and　processing,　strong　anti-interference,　non-invasive　detection,　and　are　widely　used　in　the　treatment　of　neuromuscular　diseases.　In　this　paper,　impedance　technique　is　introduced　into　grip　force　prediction.　A　single-frequency,　low-intensity　alternating　current　(AC)　signal　is　injected　into　the　brachioradialis　muscle,　and　the　change　in　muscle　impedance　is　detected　through　the　electrical　effect　of　the　electromagnetic　field　on　biological　tissue.　Then,　the　correlation　between　impedance　parameters　and　grip　force　changes　is　discussed,　and　the　Long　Short-Term　Memory　(LSTM)　grip　force　prediction　model　is　established　with　resistance　(R)　and　phase　(P)　as　feature　inputs.　The　results　show　that　the　r^{2}\_score　of　the　grip　force　prediction　model　is　greater　than　0.94　and　the　mean　square　error　(MSE)　is　lower　than　0.7.　This　paper　restores　the　actual　grip　force　based　on　the　LSTM　prediction　model　and　provides　a　new　implementation　idea　for　grip　force　prediction.　©　2022　IEEE.