Hybrid Deep Neural Network Scheduler for Job-Shop Problem Based on Convolution Two-Dimensional Transformation

Table 9

Comparison of HDNNS with other methods on traditional datasets.

		Optimal	HDNNS (our)	DQN	DEEPRM	ANN (1D)	ANN (all)	STPT	SPT
ft10; (10  10)	MAKESPAN	930	1023	1023	1025	1154	1054	1152	1169
	Scheduling score	—	90.91	90.9	90.7	80.5	88.2	80.7	79.5
	Training time	—	1804.5	13321.7	12212.2	426.1	3245.6	—	—
	Scheduling time	—	3.6	3.9	3.8	1.2	4.7	1.0	1.0
ft20; (20  10)	MAKESPAN	1165	1391	1342	1317	1524	1504	1434	1544
	Scheduling score	—	83.7	86.8	88.4	76.4	77.4	81.2	75.4
	Training time	—	3954.1	16532.1	17548.3	436.5	4689.5	—	—
	Scheduling time	—	7.6	7.4	7.2	2.4	9.2	1.9	1.9
la24; (20  10)	MAKESPAN	935	1056	1088	1071	1564	1564	1580	1569
	Scheduling score	—	88.5	85.9	87.30	59.7	59.7	59.1	59.5
	Training time	—	3976.5	16844.3	16254.5	487.6	4684.4	—	—
	Scheduling time	—	7.6	8.2	7.3	2.5	2.5	1.9	1.9
la36; (15  15)	MAKESPAN	1268	1318	1465	1465	1721	1721	1729	1729
	Scheduling score	—	96.2	86.55	86.5	73.6	73.6	73.3	73.3
	Training time	—	15318.1	63172.0	62251.4	578.5	21688.1	—	—
	Scheduling time	—	38.4	39.3	39.2	3.3	42.5	8.3	7.2
abz7; (20  15)	MAKESPAN	665	726	739	720	940	940	980	1026
	Scheduling score	—	91.6	89.9	92.3	70.7	70.7	67.8	64.8
	Training time	—	22124.2	90584.4	92584.4	683.4	29258.3	—	—
	Scheduling time	—	51.3	48.3	50.24	4.8	89.5	13.3	12.3
yn1; (20  20)	MAKESPAN	886	995	1183	1067	1183	1183	1208	1207
	Scheduling score	—	89.0	74.8	83.04	74.8	74.8	73.3	73.4
	Training time	—	30688.8	126689.2	125845.4	536.0	35648.2	—	—
	Scheduling time	—	177.2	188.0	184.5	65.4	194.5	78.4	73.5
Average	MAKESPAN	974.83	1084.83	1140.0	1110.8	1347.6	1327.6	1347.1	1374.0
	Scheduling score	—	90.01	85.5	88.0	72.6	74.1	72.6	71.0
	Training time	—	12977.7	54523.9	54449.4	524.7	16535.7	—	—
	Scheduling time	—	47.6	49.1	48.7	13.3	57.1	17.5	16.3
MAKESPAN rank			1	3	2	5	4	6	7
Training time			2	4	3	1	5	—	—
Scheduling time RANK			4	6	5	1	7	3	2