This paper studies a combination forecasting model with strong adaptability and high dimension to evaluate value of patents, and verifies the model empirically in the research. Through the AHP analysis, five necessary factors that affect the value of patents are discovered, which come from both the characteristic factors of the patent itself and the institutional characteristic factors that attempt to transform the patent. We’ve constructed a state space model through some data that can already obtain the added value of its patent through the Hejun value evaluation model and deployed the state space model in the artificial intelligence data space to have the neural network training. Another part of the known data is used to empirically verify this model, and it is found that the data fits well. Not only can the model be used for patent conversion institutions to evaluate patents, but also for patent holders to evaluate patent conversion institutions.