1.College of Electrical Engineering, Shanghai University of Electric Power, Shanghai 200090, China;2.Hubei Provincial Key Laboratory for Low-frequency Electromagnetic Communication Technology, Wuhan 430074, China;3.School of Mathematical Sciences, Shanghai Normal University, Shanghai 200234, China

For low-rise industrial buildings with poor thermal insulation performance and large surface area, the shading and insulation effect of their rooftop photovoltaic modules will have a huge impact on the magnitude and fluctuation pattern of the daily load. However, research on net load forecasting mostly focuses on the single characteristics of load and photovoltaic, with little consideration given to the thermal effects of photovoltaics on load. In response to the above shortcomings, this paper takes an industrial building in Shanghai as the research object, constructs a photovoltaic-rooftop integrated heat transfer model through the thermal balance method, calculates the annual thermal effect of rooftop photovoltaic, and verifies the strong correlation between hourly heat transfer and photovoltaic cell temperature with load and photovoltaic output through correlation analysis. Subsequently, in order to more accurately extract the behavior characteristics of the load, the paper conductes cluster analysis on the daily load of each season based on the waveform features of heat transfer. Finally, using heat transfer characteristics as input factors and bi-directional long short-term memory network as forecasting algorithm, a short-term net load forecasting method considering the thermal effect of rooftop photovoltaic is proposed. The net load data of the building in each season is forecasted, modeled, calculated, and error analyzed. Horizontal comparison with long short-term memory network-attention mechanism, long short-term memory network and extreme learning machine shows that the proposed method can significantly improve the accuracy of net load forecasting.