Fig.10 Mean absolute error rate and actual values of testing samples ranging from 1.30 p.u. to 1.36 p.u.
Fig.11 Mean absolute error rate and actual values of testing samples ranging from 1.38 p.u. to 1.44 p.u.
In addition, the 2-D nomogram of improved DT model is depicted in Fig. 12, where the splitting rules of each internal node during the DT generation process is indicated. In addition, the predicted value, sample size, and node purity of each terminal node can be obtained, achieving the visualization of the decision-making process.
Fig.12 2-D nomogram of improved DT model
Prediction effect of integrated method
(1) Theoretical analysis results
The simulation parameters of test system are as follows:U dN=±400kV; R d=6.04Ω;U N=800kV; Q CN=4000MVar;I dmin=0.5kA; the rated DC current is 5kA; the rated active power is 1000MW. Considering a typical operation scenario,Q ac=105MVar; the output of synchronous generators and renewable energy generation system is 5000MW and 1500MW, respectively. When the S C is 32000MVA, assuming that the single CF occurs in Qingyu DC, the transient overvoltage can be calculated as 1.328 according to . The simulation results are depicted in Fig. 13, which validates the effectiveness of the theoretical analysis method.
Fig.13 Voltage response curves of Qingnan bus
(2) Prediction effect of integrated method
Based on the overvoltage theoretical analysis method in section 3.1.1, the theoretical overvoltage peak values of 4480 samples generated offline are calculated and adopted as partial input features of the training samples. Along with the key electrical inputs and output labels obtained by PSASP in section 3.1.2, the sample set is formed to train the DT model, revealing the association pattern between theoretical analysis values and true values. Based on the well-trained integrated DT model, the regression prediction results are compared with traditional DT model. Results of the overvoltage peak value at Qingnan 750kV bus under different prediction models are shown in Fig. 14. It can be concluded that the integrated model proposed in this paper has a better regression prediction effect on the testing data set, and the predicted values are closer to the true values.
Fig.14 Prediction results of data-driven method and integrated method
Furthermore, performances of the traditional DT model, improved DT model and integrated method are compared in Table 3.
Table 3 Prediction effect under different evaluation indices