LTSF-Benchmark.md

January 7, 2024 · View on GitHub

Features

  • To facilitate the follow-up research, we have provided the method comparison results under commonly used datasets, and the follow-up methods will continue to be updated.
  • Support recent SOTA and traditional but popular methods.
  • Any Pull requests are welcome!

Multivariate long-term series forecasting results on four ETT datasets with best input length and prediction length O ∈{96, 192, 336, 720}

MethodsDatasetETTH1ETTH2ETTm1ETTm2
Metric96192336720961923367209619233672096192336720
LinearMSE0.3750.4180.4790.6240.2880.3770.4520.6980.3080.3400.3760.4400.1680.2320.3200.413
MAE0.3970.4290.4760.5920.3520.4130.4610.5950.3520.3690.3930.4350.2620.3080.3730.435
NLinearMSE0.3740.4080.4290.4400.2770.3440.3570.3940.3060.3490.3750.4330.1670.2210.2740.368
MAE0.3940.4150.4270.4530.3380.3810.4000.4360.3480.3750.3880.4220.2550.2930.3270.384
DLinearMSE0.3750.4050.4390.4720.2890.3830.4480.6050.2990.3350.3690.4250.1670.2240.2810.397
MAE0.3990.4160.4430.4900.3530.4180.4650.5510.3430.3650.3860.4210.2600.3030.3420.421
FEDformer-fMSE0.3760.4200.4590.5060.3460.4290.4960.4630.3790.4260.4450.5430.2030.2690.3250.421
MAE0.4190.4480.4650.5070.3880.4390.4870.4740.4190.4410.4590.4900.2870.3280.3660.415
FEDformer-wMSE0.3950.4690.5300.5980.3940.4390.4820.5000.3780.4640.5080.5610.2040.3160.3590.433
MAE0.4240.4700.4990.5440.4140.4450.4800.5090.4180.4630.4870.5150.2880.3630.3870.432
AutoformerMSE0.4490.5000.5210.5140.3580.4560.4820.5150.5050.5530.6210.6710.2550.2810.3390.433
MAE0.4590.4820.4960.5120.3970.4520.4860.5110.4750.4960.5370.5610.3390.3400.3720.432
InformerMSE0.8651.0081.1071.1813.7555.6024.7213.6470.6720.7951.2121.1660.3650.5331.3633.379
MAE0.7130.7920.8090.8651.5251.9311.8351.6250.5710.6690.8710.8230.4530.5630.8871.338
PyrafromerMSE0.6640.7900.8910.9630.6450.7880.9070.9630.5430.5570.7540.9080.4350.7301.2013.625
MAE0.6120.6810.7380.7820.5970.6830.7470.7830.5100.5370.6550.7240.5070.6730.8451.451
LogTransMSE0.8781.0371.2381.1352.1164.3151.1243.1880.6000.8371.1241.1530.7680.9891.3343.048
MAE0.7400.8240.9320.8521.1971.6351.6041.5400.5460.7000.8320.8200.6420.7570.8721.328
ReformerMSE0.8370.9231.0971.2572.62611.1209.3233.8740.5380.6580.8981.1020.6581.0781.5492.631
MAE0.7280.7660.8350.8891.3172.9792.7691.6970.5280.5920.7210.8410.6190.8270.9721.242
Linear-IMSE0.3780.4150.4470.4890.4080.5750.6431.0750.2860.3300.3720.4280.1960.2810.3800.484
MAE0.3970.4220.4410.4880.4150.5000.5430.7110.3340.3610.3870.4190.2840.3470.4060.475
DLinear-IMSE0.3770.4200.4480.4860.3990.4830.6861.0390.2860.3270.3660.4270.1820.2980.4230.799
MAE0.3960.4260.4430.4860.4090.4580.5530.7000.3350.3580.3800.4160.2740.3520.4340.580
NLinear-IMSE0.3830.4140.4380.4510.2870.3450.3710.4160.2880.3300.3710.4310.1630.2180.2730.369
MAE0.4020.4210.4360.4590.3470.3870.4130.4450.3360.3600.3830.4180.2510.2900.3260.384
MethodsDatasetElectricityExchange RateTrafficWeatherILI
Metric9619233672096192336720961923367209619233672024364860
LinearMSE0.1400.1530.1690.2030.0820.1670.3280.9640.4100.4230.4360.4660.1760.2180.2620.3261.9472.1822.2562.390
MAE0.2370.2500.2680.3010.2070.3040.4320.7500.2820.2870.2950.3150.2360.2760.3120.3650.9851.0361.0601.104
DLinearMSE0.1410.1540.1710.2100.0890.1800.3311.0330.4100.4230.4350.4640.1820.2250.2710.3381.6831.7031.7191.819
MAE0.2370.2480.2650.2970.2080.3000.4150.7800.2790.2840.2900.3070.2320.2690.3010.3480.8580.8590.8840.917
NLinearMSE0.1400.1530.1690.2030.0810.1570.3050.6430.4100.4230.4360.4660.1760.2200.2650.3232.2151.9632.1302.368
MAE0.2370.2490.2670.3010.2030.2930.4140.6010.2820.2870.2960.3150.2370.2820.3190.3621.0810.9631.0241.096
FEDformer-fMSE0.1930.2010.2140.2460.1480.2710.4601.1950.5870.6040.6210.6260.2170.2760.3390.4033.2282.6792.6222.857
MAE0.3080.3150.3290.3550.2780.3800.5000.8410.3660.3730.3830.3820.2960.3360.3800.4281.2601.0801.0781.157
FEDformer-wMSE
MAE
AutoformerMSE0.2010.2220.2310.2540.1970.3000.5091.4470.6130.6160.6220.6600.2660.3070.3590.4193.4833.1032.6692.770
MAE0.3170.3340.3380.3610.3230.3690.5240.9410.3880.3820.3370.4080.3360.3670.3950.4281.2871.1481.0851.125
InformerMSE0.2740.2960.3000.3730.8471.2041.6722.4780.7190.6960.7770.8640.3000.5980.5781.0595.7644.7554.7635.264
MAE0.3680.3860.3940.4390.7520.8951.0361.3100.3910.3790.4200.4720.3840.5440.5230.7411.6771.4671.4691.564
PyraformerMSE0.3860.3860.3780.3760.3761.7481.8741.9432.0850.8670.8690.8810.8960.6220.7391.0041.4207.3947.5517.662
MAE0.4490.4430.4430.4451.1051.1511.1721.2060.4680.4670.4690.4730.5560.6240.7530.9342.0122.0312.0572.100
LogTransMSE0.2580.2660.2800.2830.9681.0401.6591.9410.6840.6850.7340.7170.4580.6580.7970.8694.4804.7994.8005.278
MAE0.3570.3680.3800.3760.8120.8511.0811.1270.3840.3900.4080.3960.4900.5890.6520.6751.4441.4671.4681.560
RepeatMSE1.5881.5951.6171.6470.0810.1670.3050.8232.7232.7562.7912.8110.2590.3090.3770.4656.5877.1306.5755.893
MAE0.9460.9500.9610.9750.1960.2890.3960.6811.0791.0871.0951.0970.2540.2920.3380.3941.7011.8841.7981.677
Linear-IMSE0.1340.1480.1640.2010.0900.2410.3361.6580.4250.4380.4520.4820.1460.1880.2420.3172.1612.2042.2402.424
MAE0.2300.2450.2630.2970.2180.3700.4451.0330.2980.3040.3120.3300.2120.2560.3020.3591.0141.0101.0251.102
DLinear-IMSE0.1340.1480.1640.2010.0990.7820.6761.4990.4260.4380.4520.4830.1440.1870.2400.3171.9992.1422.2032.475
MAE0.2300.2450.2630.2970.2320.6390.6510.9400.2980.3040.3120.3300.2090.2540.2980.3590.9450.9771.0111.096
NLinear-IMSE0.1340.1500.1670.2060.0960.2090.3581.1330.4230.4350.4470.4760.1440.1870.2400.3181.6861.6191.6791.821
MAE0.2290.2430.2600.2930.2140.3250.4290.7850.2940.2990.3050.3220.1930.2360.2780.3330.7980.7770.8180.866

Univariate long-term series forecasting results on four ETT datasets with input length I = 96 and prediction length O ∈{96, 192, 336, 720}

MethodsDatasetETTh1ETTh2ETTm1ETTm2PaperPub.
Metric96192336720961923367209619233672096192336720
LinearMSE0.1890.0780.0910.1720.1330.1760.2130.2920.0280.0430.0590.0800.0660.0940.1200.175https://arxiv.org/pdf/2205.13504.pdfAAAI23
MAE0.3590.2120.2370.3400.2830.3300.3710.4400.1250.1540.1800.2110.1890.2300.2630.320
NLinearMSE0.0530.0690.0810.0800.1290.1690.1940.2250.0260.0390.0520.0730.0630.0900.1170.170
MAE0.1770.2040.2260.2260.2780.3240.3550.3810.1220.1490.1720.2070.1820.2230.2590.318
DLinearMSE0.0560.0710.0980.1890.1310.1760.2090.2760.0280.0450.0610.0800.0630.0920.1190.175
MAE0.1800.2040.2440.3590.2790.3290.3670.4260.1230.1560.1820.2100.1830.2270.2610.320
FEDformer-fMSE0.0790.1040.1190.1420.1280.1850.2310.2780.0330.0580.0840.1020.0670.1020.1300.178https://arxiv.org/pdf/2201.12740.pdfICML22
MAE0.2150.2450.2700.2990.2710.3300.3780.4200.1400.1860.2310.2500.1980.2450.2790.325
FEDformer-wMSE0.0800.1050.1200.1270.1560.2380.2710.2880.0360.0690.0710.1050.0630.1100.1470.219
MAE0.2140.2560.2690.2800.3060.3800.4120.4380.1490.2060.2090.2480.1890.2520.3010.368
AutoformerMSE0.0710.1140.1070.1260.1530.2040.2460.2680.0560.0810.0760.1100.0650.1180.1540.182https://arxiv.org/pdf/2106.13008.pdfNeurIPS21
MAE0.2060.2620.2580.2830.3060.3510.3890.4090.1830.2160.2180.2670.1890.2560.3050.335
InformerMSE0.1930.2170.2020.1830.2130.2270.2420.2910.1090.1510.4270.4380.0880.1320.1800.300https://arxiv.org/pdf/2012.07436.pdfAAAI21
MAE0.3770.3950.3810.3550.3730.3870.4010.4390.2770.3100.5910.5860.2250.2830.3360.435
LogTransMSE0.2830.2340.3860.4750.2170.2810.2930.2180.0490.1570.2890.4300.0750.1290.1540.160
MAE0.4680.4090.5460.6290.3790.4290.4370.3870.1710.3170.4590.5790.2080.2750.3020.321
ReformerMSE
MAE
S4MSE0.3160.3450.8250.190.3810.3320.6550.630.6510.190.4280.2540.1530.1830.2040.482https://openreview.net/pdf?id=uYLFoz1vlACICLR22
MAE0.490.5160.8460.3550.5010.4580.670.6620.7330.3720.5810.4330.3180.350.3670.567
FILMMSE0.0550.0720.0830.090.1270.1820.2040.2410.0290.0410.0530.0710.0650.0940.1240.173https://arxiv.org/pdf/2205.08897.pdfNeurIPS 2022
MAE0.1780.2070.2290.240.2720.3350.3670.3960.1270.1530.1750.2050.1890.2330.2740.323
ARIMAMSE0.0580.0730.0860.1030.2730.3150.3670.4130.0330.0490.0650.0890.2110.2370.2640.31
MAE0.1840.2090.2310.2530.4070.4460.4880.5190.1360.1690.1960.2310.340.3710.3960.441