Abstract: In midsummer extreme heat waves occur frequently, and electricity demand surges to its annual peak, posing the greatest pressure and the highest uncertainty for power supply security. To accurately forecast the midsummer peak load and enhance prediction interpretability, this paper proposes a dual-path forecasting method that integrates mechanism analysis of the "Four Effects" with time-series large models. In the mechanism analysis path, based on load composition theory, the total load is decomposed into base load and cooling load, which respectively characterize the economic growth effect, rest-day effect, temperature-rise effect, and accumulation effect. Historical data of the summer peak load in 2025 are reviewed and decomposed, and then each effect component for midsummer 2026 is extrapolated to obtain the peak load forecast. In the data-driven path, the Chronos-2 time-series large model is adopted, integrating multi-source information including economic, meteorological, and calendar data for daily peak load forecasting. It is compared with patch time series transformer and light gradient boosting machine algorithms to select the best performer, and the mechanistic and data-driven paths are cross-validated against each other. The results show that, considering the latest economic and climatic conditions, the maximum load of a certain regional power grid in midsummer 2026 is 1305 GW, consisting of a base load of 934 GW (economic growth effect 934 GW, rest-day effect 0 GW) and a cooling load of 371 GW (temperature-rise effect 340 GW, accumulation effect 31 GW). The time-series large model performs best on the test set, with a mean absolute percentage error of only 1.64%, and forecasts a peak load of 1306 GW, which is highly consistent with the conclusion of the "Four Effects", thus verifying the reliability of the proposed method.