Renewable energies are nowadays one of the better solutions for environmental reasons and to meet consumer demand. Solar energy is attracting widespread attention from countries around the world, and increasing its effectiveness is becoming a challenging issue. In this manuscript, a double mode maximum power point tracking (MPPT) algorithm is presented. A radial basis function neural network (RBFNN) is designed and driven by history temperature and irradiance data to realize PV MPPT in both normal operation and partial shading condition (PSC), while the traditional perturb and observe (P&O) method is adopted to realize MPPT under normal condition. The occurrence of the partial shading condition is detected by Long short-term memory (LSTM) based solar power prediction, Deviation between the predicted and real measured power is used to trigger the MPPT modes switching smoothly and adaptively, thus the fastness and accuracy of MPPT are improved with uncertain solar irradiance and ambient temperature. To prove the validity of the presented MPPT method, a PV system model is built on MATLAB/Simulink. Serval cases including normal operation and PSC are considered. The comparison results with existing traditional MPPT methods indicate the superiority of our method. It can accurately diagnose the occurrence of PSC and quickly realize MPPT under various working conditions.