The output power of photovoltaic (PV) power sources is significantly affected by factors such as weather and light intensity, exhibiting notable intermittency and volatility. When the grid-connected capacity of PV is large, these characteristics cause the power load to exhibit non-stationary random characteristics, thereby increasing the difficulty of load forecasting. Traditional load forecasting methods neglect the intermittency and volatility of PV power sources. As a result, when dealing with complex load time series, they struggle to accurately calculate the correlation coefficients, leading to large forecasting errors. To address this issue, this paper proposes an ultra-short-term power load forecasting method that takes into account the intermittency and volatility of PV power sources. Historical power load data is collected to extract the load change patterns. Meanwhile, the real-time operating data of PV power sources is gathered to calculate the intermittency coefficient and volatility coefficient, and to clarify the influence relationship between these coefficients and the power load. Then, based on the time interval between the ultra-short-term forecasting time node and the current node, an initial forecast value of the power load is generated. The initial forecast value is corrected using the calculation results of the above-mentioned influence parameters to obtain an accurate forecast result. Performance test experiments demonstrate that under the two operating conditions of normal days and rest days, the forecast values obtained by this method are highly consistent with the actual power load values, demonstrating obvious application advantages.