Cost evolution has an important influence on the commercialization and large-scale application of power technology. Many researchers have analyzed the quantitative relationship between the cost of power technology and its influencing factors while establishing various forms of technical learning curve models. In this paper, we focus on the positive effects of the policy on research and development (R&D) learning by summarizing and comparing four energy technology cost models based on learning

more »

... rves. We explore the influencing factors and dynamic change paths of power technology costs. The paper establishes a multi-stage dynamic two-factor learning curve model based on cumulative R&D investment and the installed capacity. This work presents the structural changes of the influencing factors at various stages. Causality analysis and econometric estimation of learning curves are performed on wind power and other power technologies. The conclusion demonstrates that a "learn by researching" approach had led to cost reduction of wind power to date, but, in the long term, the effect of "learn by doing" is greater than that of "learn by researching" when R&D learning is saturated. Finally, the paper forecasts the learning rates and the cost trends of the main power technologies in China. The work presented in this study has implications on power technology development and energy policy in China. Sustainability 2017, 9, 861 2 of 14 reduction, which better reflected the endogenous changes in technology. The learning rate is associated with the rate of decline of technology costs within a certain period. Along with the deepening and development of the research, some researchers believed that the single-factor learning curve for energy technology had big limitations. Ref. [2] only considered the extent of demand for the technological learning progress based on the single-factor learning curve, without consideration of the impacts of supply, leading to a certain degree of deviation about learning rate forecasts. Ref. [3] studied wind power technology by an alternate learning curve model, and thought that the learning rate, predicted by the single-factor learning curve model, was extremely sensitive to statistics and explanatory variables, which means cautious when using this model. In recent years, the study of the learning curve has been extended from single-factor equations to two-factor equations, which considered the effect of learning by researching. Ref. [4] first put forward that the cumulative R&D and production were two factors driving the technological cost reduction. Refs. [5-8] measured the learning rate of energy technology and power generation technology via a two-factor learning curve, and analyzed the technology cost trends by learning from experience and R&D. Through in-depth analysis of these case studies, the learning rate of "learn by researching" was more than the corresponding vector of "learn by doing". This study demonstrated that strengthening R&D investments was more helpful to improving technical performance, with respect to increasing the use of technology, while some researchers believe that the single-factor equation missed an important variable named R&D investment, by comparing the estimation learning rate of single-factor and two-factor equations, so the learning rate of "learn by doing" was overestimated, causing information distortion in a certain degree. Foreign studies have shown that the investment cost of wind power gradually declined with the accumulation of knowledge and experience. The development of wind power technology was in line with the learning curve model [9] . In addition, the development of photovoltaic power generators and other new energy technologies was also in line with the learning curve model [10, 11] . Therefore, the generic quantitative analysis of renewable energy technology was the learning curve model. Ref. [12] analyzed wind turbines, solar photovoltaic modules and other renewable energy technology innovation solutions by using learning curves. The study found the likelihood of renewable energy technologies to reduce costs was much higher than traditional energy technologies. To make the wind power and PV power more economical, compared with the conventional power plant, it could take a large upfront investment. Government policy and other methods for achieving a significant decline in technology cost, and accelerated the promotion of renewable energy technologies. Ref. [13] used the learning curve and the Grey Model (GM) (1, 1) method to predict the change trend of wind power cost, and analyzed the main factors affecting the wind turbines power efficiency, and put forward the corresponding wind power pricing schemes. In China, the learning curve research of energy technology also had a certain degree of accumulation. Ref. [14] used the learning curve and the GM (1, 1) model and analyzed the investment cost change trend of wind power in China. Ref. [15, 16] used the learning curve model to compare the wind power cost in the condition of constant learning rate and changed the learning rate in stages. Ref. [17] analyzed the timing and scale prediction of wind power and PV power accessed via a smart grid based on the learning curve. Ref. [18] adopted the learning curve model to calculate the technological learning rate based on the different lowest price scenarios of PV components, and analyzed the relationship between the learning rate and the lowest price. Ref. [19] established the two-factor learning curve model for solar PV power generation technology, based on the Wright basic learning curve, and deeply analyzed the cumulative production and cumulative R&D investment influenced on the power generation cost of solar PV. Refs. [20, 21] made the endogenous technology learning module, coupled to the energy system optimization model MARKAL, and researched sustainable energy development and utilization in the western China. Ref. [22] studied the development path and emission reduction cost of wind power and carbon capture technology based on the learning curve, and discussed the two types of technology that could instigate market competition and the emission reduction potential and cost. Ref. [23] constructed the single-factor