Building thermal load small sample prediction method based on simulation data and meta-learning method

The invention relates to a building thermal load small sample prediction method based on simulation data and a meta-learning method, and the method comprises the steps: determining the geometric shape of a simulation building according to the building type of a target building, setting the probability distribution of building parameters, and carrying out the Monte Carlo sampling, and obtaining different building parameter sets representing different use conditions of the building; performing simulation to obtain thermal load data of the building under different use conditions as source building data; constructing each group of source building data into an independent meta-task, and training on the meta-task by using a meta-learning method to obtain a thermal load small sample prediction basic model suitable for the same type of buildings; and for a target building with insufficient data, performing parameter initialization on the target task model by using the obtained parameters of the basic model, and perfo