**4. Validation**

In order to verify the effectiveness of the proposed CPM method, its prediction result is compared with PM, CM and real value. MRE is used to evaluate the performance of the prediction model.

The proposed method to predict the electricity consumption of SETS is verified using real operation data, the data is obtained from the power grid valley thermal storage system monitoring platform of State Grid Company. This paper uses the data of one customer from the monitoring platform. The SETS is located in the third printing plant of Heping District, Shenyang City, Liaoning Province. The scene device of SETS is shown in Figure 10. The nominal power of the SETS is 1 MW, and the maximum thermal storage capacity is 10 MWh.

**Figure 10.** Scene device of SETS.

The load data of 1MW SETS during the heating period from 10 November 2017, to 19 March 2018, are taken. The PM parameters of SETS are selected as shown in Table 2. The data of 120 days are used as the training set, and the data of the later 10 days are used for the testing. The simulation results are shown in Figure 11. The four curves are the individual SETS real values, CPM, PM, and the CM predicted value. The trained model is used to test the load condition in the following 10 days. From the training results, the CPM predicted value coincides with the real value. In the testing results, the comparison data in the orange dotted frame is enlarged for 121–130 days as seen in Figures 12–14. The result after the application of the CPM is significantly better than that individually predicted by the PM and CM alone.


**Table 2.** PM parameters.

**Figure 11.** The simulation results of one customer's 130 days.
