Abstract:    As an important subject of innovation, the innovation ability of enterprises is reflected in all the patents it owns, usually in the form of patent portfolio. It is of great significance to accurately evaluate the value of patent portfolio. In the macro sense, it is the necessary guarantee for the State to measure the innovation ability of enterprises and formulate reasonable industrial technology policies. In the microcosmic sense, it is the pre-condition for the smooth progress of enterprise technical cooperation, transaction and pledge financing, and it is also the need of product and project investment evaluation.
   Based on the compound two-tier patent portfolio as the research object, the paper selects the indicator of forward cites, claims, family, backward cites and so on to construct the value evaluation model. First of all, the statistically relevant method is used to combine similar indicators of the value of patents within similar patent portfolios, and then to design a variety of patent grouping structures, to analyze the optimal grouping structure of patent groups by structural equations (SEM), and then to select the optimal variable groups and important variables using the group exponential lasso method (GEL), so as to determine the pricing formula of patent portfolio.
    After analyzing the drug as an example, it is found that the combination of internal indicators of similar patent groups is better by geometric average, and the combination between patent groups is better by patent category, and the formula of drug patent portfolio value is obtained. The main conclusions are: compound patent determines the value of patent portfolio, product patent plays a complementary role, other types of patents have little effect on the portfolio value itself; in patent indicators, forward cites indicators are the most important, claims, the family plays a complementary role, backward cites do not have a significant role.
   Later, the SEM-GEL method proposed in this paper is compared with the author′s other methods of further innovating and expanding the design according to traditional ideas, it shows that it has the advantages of reasonable grouping basis, high fit degree and clear combination form.
    In addition to analyzing the complex two-tier patent portfolio of drugs, the model can be used to analyze the complex multi-patent portfolio of different industries and other issues，such as patent portfolio financing screening, high-tech product technology value evaluation, technology enterprise research and development strength measurement, research and development performance evaluation of researchers and many other applications and research fields.
     The innovation point of the thesis is as follows: (1) the quantitative evaluation of complex double-layer patent portfolio is studied. The existing research only involves a single-layer patent combination, and most of the research is qualitative or simple related analysis; this paper considers the interaction between the internal patent groups of the portfolio, as well as the interaction between the patents within the group, and conducts a quantitative analysis. (2) Use the competitive analysis of structural equation model to obtain the optimal grouping form of patent. Existing studies on the interaction of patent values within patent portfolios are only theoretical analyses. However, competitive analysis can be used to verify the theoretical inference of the interaction between patent groups. (3) The group exponential lasso method is selected to determine the pricing formula of the patent portfolio. Using this method, different groups of variables can be filtered, and internal variables of the same group of variables can also be filtered. The overall impact of each group on the explained variable can be compared, so that the key variable groups and key variables can be clearly identified, which provides strong empirical support for the analysis of the problem.

Key words:  patent portfolio value, portfolio structure analysis, compound double-layer portfolio, structural equation model, group exponential lasso