*3.4. CSO-ELM Model*

In the basic ELM model, when the number of hidden layer nodes is fixed, the input weights and hidden layer deviations of the network structure are randomly determined, which affects the stability of the model itself. After using CSO to improve ELM algorithm, the number of hidden layer nodes, input weights and thresholds value of ELM model are adaptively selected. The flow chart is shown in Figure 1. In the CS algorithm, *N* nests are set. Then, round the position value as the number of the hidden nodes *M*. After inputting the sample data, a set of input weight matrices *W* for ELM networks will be generated. Then, the threshold value of the hidden layer *b* will be obtained, and the model output the weight matrices β. The prediction error calculated by the sample data will be the fitness value of the present nest. At the end of each iteration, the optimal nest location is preserved, and the values of *W*, *b*, and β are outputted. After satisfying the iteration termination conditions, the fitness values of nests are compared. Finally, the optimal position of the bird's nest are obtained, and the corresponding values of *W*, *b*, β will be exported. The improved ELM method operates as follows.


**Figure 1.** Outline of procedure of proposed method.

### **4. Case Study and Results**

#### *4.1. Validation Test of Forecasting Model*

The proposed forecasting method is tested by historical data inputting. The data used in this research is collected from both the "Statistical Yearbook of China" and the "Annual Report of Electric Power Industry" for the latest 20 years. Given the development of electric power substitution projects being affected by many factors, Pearson correlation test is carried out for test the correlations of each factor. Factors with high correlation are used as input data of the forecasting model to accurately predict the market potential of electric power substitution projects. The historical data are shown in the Table 3. The correlation test of each factor is carried out by Statistical Product and Service Solutions (SPSS) software and the results are shown in Table 3 and Figure 2.



**Figure 2.** Pearson coefficient.

The research is used to forecast the market potential of power substitution projects in 2018–2030. A forecasting model of extreme learning machine based on the cuckoo search optimization is constructed. By taking historical data of 1998–2017 as an example, the validity of this model is tested. Due to the substitution work being a gradual process—when the CSO-ELM model is used to forecast the market potential of electric power substitution projects—the electric power substitution amount in every year in the future is forecasted in turn. When forecasting, the data of the target year is added to the training set of the model to give out more accurate results.

Data from 1998 to 2012 is chosen as the training set, and data from 2013 to 2017 is used as the test set. The BP neural network, ELM and CSO-ELM forecasting models are all used to forecast the electric power substitution from 2013 to 2017. ELM and BP neural network are introduced as the contrast algorithm, and the parameters are set to be at the optimal values after comparison. In setting the parameters settings of BPNN, the number of neurons in hidden layer is set as 6. In addition, the tansig function is used as the transfer function. The output layer is set as the purelin function. The training times are set to 1000, and the precision target is set as 0.001. In the ELM forecasting model, the number of nodes in hidden layer is 30, and the activation function is sig function. The initial population number of CSO-ELM is 20 and the maximum iteration number is 200. Historical data is listed in Table 4.



The fitting results of the three methods are shown in Figure 3. The overall fitting effect of ELM is better than that of BPNN, which is mainly because the forecasting effect of BPNN largely depends on the quality of historical data and the training of a large number of data. The ELM method is an improved BPNN, which maintains the advantages of the fast learning speed and strong generalization ability, and can accurately analyze in the case of small amount of data. The CSO-ELM method combines the global optimization ability of CSO and optimizing transmission weight of ELM, contributing to a better forecasting effect.

**Figure 3.** Results of three forecasting models.

To quantify the effect of the forecasting model, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE) and Determining Factor *R*<sup>2</sup> are selected as indicators to evaluate the effect of the forecasting model. The calculation of the four error indicators are as follows.

$$RMSE = \sqrt{\frac{1}{n} \sum\_{i=1}^{n} (p\_i - p\_i')^2} \,\tag{15}$$

$$MAPE = \frac{1}{n} \sum\_{i=1}^{N} \left| \frac{p'\_i - p\_i}{p\_i} \right| \times 100\% \tag{16}$$

$$MAE = \frac{1}{n} \sum\_{i=1}^{N} |p'\_i - p\_i| \tag{17}$$

*R*<sup>2</sup> = *n i*=1 *p <sup>i</sup>* − *pi* 2 *n i*=1 *pi* − *pi* <sup>2</sup> , (18)

where, *pi* represents the actual electric power substitution amount. *p <sup>i</sup>* is the forecasting amount. *pi* is the average of the actual electric power substitution amount, and *n* is the data size. The calculation results of four indicators are shown in Table 5 and the contrast is shown in Figure 4.

**Table 5.** Calculation results of error indicators.


**Figure 4.** Errors of three methods.

From Table 5 and Figure 4, the constructed CSO-ELM model has a better fitting effect, higher prediction accuracy and less error in the market potential forecasting of electric power substitution. Therefore, using CSO-ELM model to forecast the substitution potential in the next 13 years can reflect the future market potential of electric power substitution projects.

#### *4.2. Scenarios Setting*

In forecasting the developing potential of future electric power substitution projects, factors such as the costs, policy support and subsidy mechanism are fully considered. Thus, combined with the political objectives of energy-consumption structure optimization, energy conservation and emission reduction, and renewable energy development, four scenarios are designed: basic scenario, high-cost restraint scenario, policies-supported scenario and subsidy weakening scenario.

In basic scenario, according to the development trend of power substitution related factors in 1998–2018, the market development potential of electric power substitution in 2019–2030 is forecasted. In the high-cost restraint scenario, the substitution project will be slowed down by the purchase of expensive equipment and high operating costs. In the policies-support scenario, in order to reduce the cost resistance and alleviate the environmental pressure caused by fossil energy combustion, the government encourages the substitution work by policy mechanism. China's "13th Five-Year Plan for Electric Power Industry" proposed to encourage the developing of electric power substitution and clean energy substitution to expand the proportion of electricity in energy consumption. In addition, the substitution work is supported with appropriate subsidies mechanisms. Relevant subsidy mechanisms can stimulate the rapid development of electric power substitution projects. With the popularization of electric power substitution, the cost would be gradually recovered, and the projects may be profitable, which is when the subsidy mechanism can be weakened or cancelled accordingly, to achieve an independent development of the substitution projects.

In addition, considering the stage characteristics of social and economic development, the forecasting period is divided into three stages: the first stage (2019–2020), the second stage (2021–2025) and the third stage (2026–2030). The parameter settings for each scenario and time period are shown in the Table 6 below.


will slow down to about 3.2% to

4.3%.

 **6.** Scenarios and primary parameter settings.

**Table**



5.2% to 6.1%. According to the carbon

While high cost brings resistance

to the

energy, it will also weaken the

environmental

reduce the e emission reduction indirectly. The trend of emission reduction

is about 1.4% to 1.7%.

 benefits and fficiency of carbon

development

 of renewable

The renewable energy will

effectively reduce carbon

emissions. Therefore, in this

scenario, the reduction rate

of carbon emissions is about

3.6% to 4%.

development

 of The decreasing rate is

slightly lower than that in

policies-supported

at about 3.6% to 3.8%

 scenario,

emission situation in the

past 20 years, assume that

the trend of carbon emission

reduction in the future is

about 2.1% to 2.5%.

**Carbon Emissions**

### *4.3. Results of Scenarios and Discussion*

In the four different scenarios, the amount of electric power substitution shows significant growth trends before 2030. Affected by electricity consumption, renewable energy generation and other factors, the market potential of electric power substitution has broad market prospects. Electric energy can effectively reduce carbon emissions to a minimum through scale-effect and technological means (such as smart grid) in power production and transmission section. Therefore, the realization of electricity substitution is a low-carbon energy development and utilization strategy as a whole, which will inevitably have a positive impact on China's low-carbon economy. Restricted by high construction costs and operation fees, people's subjective acceptance of electric power substitution project is relatively low, which brings difficulties in popularizing. The forecasting results are shown in Table 7.


**Table 7.** Multi-scenarios forecasting result.

In all four scenarios, substituting electricity shows significant increase. In the first stage, the growth rate is relatively small. During this period, the subsidies of the subsidy weakening scenario have not been weakened, so there is little difference with other scenarios. In the second stage, due to social development and technological progress, electric power substitution projects begin to have a certain scale of promotion. The substitute electricity in the high-cost restraint scenario shows the slowest increase. In the policies-supported scenario, the substitute electricity amount shows accelerated growth, while the subsidy mechanism in the subsidy weakening scenario begins to fade down slightly. In the third stage, after the promotion of the first two stages, the electric power substitution project has a certain scale effect, and the electric power substitution quantity shows a trend of accelerating growth.

As can be seen in Figure 5, the process of electric power substitution in the policies-support scenario and subsidy weakening scenario is significantly higher than that in the basic scenario, which shows that the government support is very important in substitution promoting work. In addition, the substitution amount in the high-cost restraint scenario is the least one in all four scenarios, about 538.17 million tons of standard coal in 2030. High costs lower the residents' acceptance of projects, and the promotion power of manufacturers is insufficient without any subsidies given. In the policies-support scenario, the government supports by giving subsidies in equipment purchasing, installation, operation and maintenance process—so that the projects are promoted—with the most increment in electricity amount, reaching 693.8 million tons of standard coal in 2030. Differing from the policies-support scenario, changes after power substitution reaching certain scale effects are considered in the subsidy weakening scenario. With the popularization of electric power substitution projects and technical progress, government support gradually reduced to raise independence of the substitution industry, as can be seen from the curve of the subsidy weakening scenario in Figure 5. The increasing trend of substitute electricity under the subsidy weakening scenario is similar to that of the policies-support scenario. With the reduction of subsidies, the amount of electric power substitution become less than that in the policies-support scenario after 2023, but still higher than that in the basic scenario, which shows the effectiveness of subsidy weakening.

**Figure 5.** Forecasting results of electric power substitution under all scenarios (million tons of standard coal).

In the early stage of developing electric power substitution, the main fields for promotion are concentrated on substitutions of coal, coal-fired boilers and coal-fired heating—where technology is relatively mature. In the middle and later stages, electric power substitution work turns to household electrification, substitution to internal-combustion engines by electricity and substitution to oil-fired vehicle by electric vehicle. Implementing electric power substitution will bring significant changes in China's energy consumption structure, significantly reducing the proportion of oil and natural gas consumption and significantly reduce carbon emissions.

#### **5. Conclusions**

Electric power substitution is important in environmental management and for optimizing energy structure. Market forecasting on electric power substitution projects is conducive to the government and society to develop and promote the substituting work. Therefore, a market potential forecast model of electric power substitution based CSO-ELM method is proposed. Firstly, the validity of the influencing factors on the potential of power substitution development are verified through correlation test. Secondly, combined with CSO algorithm, the ELM method is improved, which overcomes the limitation of the search ability of the original ELM and improves the forecasting accuracy. In addition, compared with BP, ELM and other forecasting models, the proposed CSO-ELM model has lower error and deviation, showing better forecasting effect. Due to government support and the limitations of industrial development, four scenarios were designed to give specific forecasting results to show how different factors influence the development of electric power substitution projects. In addition, the forecasting time period was divided into three stages to take on further analysis. Through comparative analysis of the forecasting results under different scenarios, some suggestions for promoting electric power substitution are proposed below.


**Author Contributions:** Conceptualization, J.W. and G.D.; methodology, K.W.; software, K.W.; validation, G.D., Z.T. and L.P.; formal analysis, J.W.; investigation, Q.T.; resources, L.J.; data curation, K.W.; writing—original draft preparation, J.W.; writing—review and editing, G.D. and Q.T.; K.W.; supervision, Z.T.; project administration, G.D. and L.J.; funding acquisition, Z.T. and L.J.

**Funding:** This work was partially supported by Supported by the Project funded by China Postdoctoral Science Foundation (2019M650024), the National Nature Science Foundation of China (Grant Nos. 71904049,71874053, 71573084), the Beijing Social Science Fund(18GLC058) and the 2018 Key Projects of Philosophy and Social Science Research, Ministry of Education, China (18JZD032).

**Acknowledgments:** Teachers and classmates helped complete this paper. We would like to express our gratitude to them for their help and guidance.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
