Rolling Optimization of Renewable Dispatch next, an mctS process similar to the one in alphaGo can be employed. as shown in Figure 9, a complete cascading outage path begins from the root node (the initial state) to the leaf node (the terminal). the cascading outage prediction takes the forward tree traverse from the root until a leaf node is reached. then, the risk indices of the states on this certain path are updated via a reversed tree traverse from a terminal to the root. the above process can be repeated for every outage path starting from the current state. Finally, the state with the highest risk index is selected as the most critical state that requires immediate corrective or preventive measures. rolling optimization has been used as an efficient method to accommodate stochastic renewable energy integration into the bulk power system. the rolling optimization process can be briefly described as follows: at each time interval, a whole-cycle generator dispatch schedule is calculated based on the present operating conditions as well as the predicted future conditions. only the first-hour dispatch of the schedule needs to be implemented. then the time window moves to the next interval, and the above calculation is repeated. Since forecast error decreases as it comes closer to real G 3 4 11 6 5 113 9 10 31 29 28 G 27 114 37 15 13 34 35 16 8 17 30 73 G 20 32 115 G 24 22 G 74 G 84 85 86 G 63 60 64 G 67 G 68 G 59 51 50 49 70 72 58 57 48 46 47 71 23 25 G 56 55 54 52 44 43 45 38 19 18 26 53 33 14 7 41 42 39 G 12117 69 116 66 G 61 62 65 79 81 G 99 78 75 80 108 77 98 97 96 82 G 104 106 83 95 94 100 G 107 93 108 WF7 88 89 92 G 109 101 102 103 90 110 91 G 112 87 G 111 118 76 10 7.5 4 11 6 5 113 2.5 9 0 10 31 29 28 G 114 27 -2.5 37 15 13 34 35 16 8 17 30 32 115 25 24 22 85 86 DNN Classification G 64 G 69 116 67 66 G 61 62 65 79 81 G 99 78 80 108 77 98 97 96 82 G 104 106 83 95 94 100 G 107 93 108 WF7 88 89 92 G 109 101 102 103 90 110 91 G 87 112 G 111 75 84 G 68 G G -10 G 70 74 59 63 60 51 50 49 71 72 58 57 48 46 47 G -7.5 Extract Current Feature 44 73 G 20 23 -5 56 55 54 52 43 45 38 19 18 26 53 33 14 7 41 42 39 G 12117 3 5 40 2 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 40 2 1 118 76 Locate System Fault figure 8. Fault diagnosis based on a DNN classification. RI(21) Outage Path Pr(21) Risk Index Update Pr(11) Pr(22) 31 RI(31) Pr(32) 32 RI(32) 33 RI(33) 34 RI(34) Pr(35) 35 RI(35) Pr(36) 36 RI(36) Pr(37) 37 RI(37) 21 RI(11) 11 Pr(31) RI(22) 22 Pr(33) Pr(34) RI(23) 0 Pr(23) Current State 23 RI(24) Pr(24) Pr(12) 12 RI(12) Primary Contingency 24 RI(25) Pr(25) 25 Secondary Contingency Tertiary Contingency figure 9. The MCTS for cascading outage prediction in a Markovian tree. march/april 2018 ieee power & energy magazine 83