Abstract: With the normalized operation of China’s electric?power spot market, day?ahead electricity price acts as the core variable of market clearing. Its accurate prediction provides crucial support for revenue management of market participants and day?ahead supply?demand regulation of power grids. Day?ahead electricity price is affected by coupled nonlinear factors including intra?day load fluctuation, stochastic wind?solar power output and time?varying supply?demand patterns. Obvious distribution heterogeneity exists between valley?period and peak?period electricity prices. In valley periods, sufficient power supply results in stable and convergent electricity?price values. By contrast, surging load and scarce peak?shaving resources in peak periods easily trigger extreme price spikes. Global machine?learning models with unified parameters fail to adapt to such differentiated data distributions, which leads to excessive prediction errors during peak hours.