Academia.eduAcademia.edu

Figure 1 – uploaded by Yang Li

See full PDFdownloadDownload figure
Fig. 1. Flowchart of feature selection based on MA termination condition is met; otherwise, the evolutionary generation g is increased to g +1 and go to Step 4. 4. Construction of the original feature set In this paper, the original feature set is constructed based on the following principles:

Figure 1 Flowchart of feature selection based on MA termination condition is met; otherwise, the evolutionary generation g is increased to g +1 and go to Step 4. 4. Construction of the original feature set In this paper, the original feature set is constructed based on the following principles:

Related Figures (5)

Fig. 2. New England 39-bus test system 5.1.1. Generation of the sample sets The New England 39-bus power system is a well-known test system for TSA studies reported in the literature [10-13, Extensive time domain simulation work has been carried out to create the training and test sample sets. The
Fig. 2. New England 39-bus test system 5.1.1. Generation of the sample sets The New England 39-bus power system is a well-known test system for TSA studies reported in the literature [10-13, Extensive time domain simulation work has been carried out to create the training and test sample sets. The
Training results of different optimization algorithms From Table 2, it can be seen that all the optimization algorithms can effectively complete the feature selection task Using the proposed feature selection approach, an optimal feature subset {Tz1, Tz4, Tz19, Tz24, Tz26, Tz31, Tz33} (named as Aj:) is extracted from the original feature set. At the same time, a comparative test is carried out using some other algorithms, such as genetic algorithm (GA), discrete particle swarm optimization (DPSO) and BDE. The common parameters of the MA algorithm and the three other algorithms, such as population size and maximum number of iterations, are given the same values, and the other parameters are set as follows: crossover probability and mutation probability 0.85 and 0.01 respectively in GA, and the learning factor c, =c, = 2, inertia weight @=1 in DPSO. Considering the randomness of intelligent optimization algorithms, all the algorithms are independently run 100 times, and the results are shown in Table 2, where the search time is the average search time of 100 times. Tahkhla 3?
Training results of different optimization algorithms From Table 2, it can be seen that all the optimization algorithms can effectively complete the feature selection task Using the proposed feature selection approach, an optimal feature subset {Tz1, Tz4, Tz19, Tz24, Tz26, Tz31, Tz33} (named as Aj:) is extracted from the original feature set. At the same time, a comparative test is carried out using some other algorithms, such as genetic algorithm (GA), discrete particle swarm optimization (DPSO) and BDE. The common parameters of the MA algorithm and the three other algorithms, such as population size and maximum number of iterations, are given the same values, and the other parameters are set as follows: crossover probability and mutation probability 0.85 and 0.01 respectively in GA, and the learning factor c, =c, = 2, inertia weight @=1 in DPSO. Considering the randomness of intelligent optimization algorithms, all the algorithms are independently run 100 times, and the results are shown in Table 2, where the search time is the average search time of 100 times. Tahkhla 3?
Fig. 3. Fitness curves of different optimization algorithms shown in Fig. 3. From Table 2, it can be seen that all the optimization algorithms can effectively complete the feature selection task. Since the optimization mechanisms of the algorithms are different, the proposed approach and BDE have better performance than the other two algorithms, such as higher fitness value, shorter search time and higher search success rate. At the same time, MA adopts an adaptive adjustment mechanism for control parameters and the “coarse and fine’ search strategy, thus it has the best result and the most stable performance. In the optimization process, the fitness value evolution curves for the four different optimization algorithms are
Fig. 3. Fitness curves of different optimization algorithms shown in Fig. 3. From Table 2, it can be seen that all the optimization algorithms can effectively complete the feature selection task. Since the optimization mechanisms of the algorithms are different, the proposed approach and BDE have better performance than the other two algorithms, such as higher fitness value, shorter search time and higher search success rate. At the same time, MA adopts an adaptive adjustment mechanism for control parameters and the “coarse and fine’ search strategy, thus it has the best result and the most stable performance. In the optimization process, the fitness value evolution curves for the four different optimization algorithms are
Test results on the New England 39-bus power system As shown in Table 3, compared with the original feature set A, A; has similar classification accuracy, but the data dimension is reduced to 7 from 33. Meanwhile, although the proposed method selects the less features than PCA, its classification performance is better than the latter. At the same time, it can be seen that the three TSA models have similar classification performance using A; and A, therefore the optimal feature subset obtained has a good suitability for different TSA models. As shown in Table 3, compared with the original feature set A, A; has similar classification accuracy, but the data The test results of stability classification are given in Table 3. indicator 7 is used to comprehensively evaluate the TSA classifier models’ performance. 77 is defined as
Test results on the New England 39-bus power system As shown in Table 3, compared with the original feature set A, A; has similar classification accuracy, but the data dimension is reduced to 7 from 33. Meanwhile, although the proposed method selects the less features than PCA, its classification performance is better than the latter. At the same time, it can be seen that the three TSA models have similar classification performance using A; and A, therefore the optimal feature subset obtained has a good suitability for different TSA models. As shown in Table 3, compared with the original feature set A, A; has similar classification accuracy, but the data The test results of stability classification are given in Table 3. indicator 7 is used to comprehensively evaluate the TSA classifier models’ performance. 77 is defined as
From Table 4, it can be observed that B, has similar classification performances with A, but the data dimension is reduced to 11 from 33. The classification performance of B; is better than Bz. Furthermore, the composite indicator 7 of B; is higher around 2.0% than A;. This shows that as the size of the systems increases, complexity of the transient stability pattern increases, and the number of selected input features should also be increased accordingly to provide more adequate knowledge to fully reflect the power system stability characteristics. Test results on the southern power system of Hebei province
From Table 4, it can be observed that B, has similar classification performances with A, but the data dimension is reduced to 11 from 33. The classification performance of B; is better than Bz. Furthermore, the composite indicator 7 of B; is higher around 2.0% than A;. This shows that as the size of the systems increases, complexity of the transient stability pattern increases, and the number of selected input features should also be increased accordingly to provide more adequate knowledge to fully reflect the power system stability characteristics. Test results on the southern power system of Hebei province
Related topics:
Tabu SearchRough SetsMemetic AlgorithmsKernel MethodsPMU based monitoring and controlDifferential Evolution AlgorithmTransient Stability of Power SystemsPower System Modeling and SimulationPhasor Measurement units
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved