Traffic-speed-prediction

We have added all the baseline models‘ results and codes to our GitHub page, including ST-GRAT, GMAN, AST-GAT, DCRNN, T-GCN, Etc! Note, we have not added the ST-GRAT and DCRNN results to our paper because the prediction precisions of DCRNN and ST-GRAT (implemented by PyTorch, torch==1.6.0) are further lower than GMAN, but we have gave the results on the dataset as the following shows. If you need, we will add both results to this paper!!! Our task is using the 6 historical time steps to prediction the future 6 target time steps traffic speed of highway, and our time granularity is 15 minutes. If you have any questions, don't hesitate to connect us, thanks！

注意事项

需要注意的是，需要根据requirements.txt文件中指示的包进行安装，才能正常的运行程序！！！

首先，使用conda创建一个虚拟环境，如‘conda create traffic_speed’；

激活环境，conda activate traffic_speed；

安装环境，需要安装的环境已经添加在requirements.txt中，可以用conda安装，也可以使用pip安装，如：conda install tensorflow==1.12.0；

如果安装的是最新的tensorflow环境，也没问题，tensorflow的包按照以下方式进行导入即可：import tensorflow.compat.v1 as tf tf.disable_v2_behavior()；

点击 run_train.py文件即可运行代码。

需要注意的是，我们在tensorflow的1.12和1.14版本环境中都可以运行

Experimental Results

HA (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	6.185242	6.185242	6.185242	6.185242	6.185242	6.185242
RMSE	10.139710	10.139710	10.139710	10.139710	10.139710	10.139710
MAPE	0.126192	0.126192	0.126192	0.126192	0.126192	0.126192
R	0.941854	0.941854	0.941854	0.941854	0.941854	0.941854
R²	0.886221	0.886221	0.886221	0.886221	0.886221	0.886221

ARIMA (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.883667	6.046000	6.104512	6.374540	6.360256	6.168883
RMSE	9.429440	9.798609	9.769539	10.209594	10.030069	9.769304
MAPE	0.102238	0.159162	0.130914	0.110670	0.129478	0.131120
R	0.941224	0.937721	0.937789	0.930736	0.933817	0.937160
R²	0.885876	0.879284	0.879403	0.866113	0.871992	0.878266

SVM (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.622419	5.694858	5.636928	5.902129	5.898720	5.929169
RMSE	9.541193	9.459830	9.366811	9.714313	9.744125	9.761787
MAPE	0.111848	0.114599	0.126784	0.138657	0.142691	0.127152
R	0.948368	0.949696	0.950375	0.947028	0.946775	0.946549
R²	0.898497	0.901055	0.902324	0.895883	0.895211	0.894915

LSTM (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.517776	5.632012	5.777946	5.946635	6.041769	6.079495
RMSE	9.125115	9.284576	9.433199	9.641906	9.626469	9.586535
MAPE	0.133428	0.140966	0.126444	0.119835	0.122028	0.134512
R	0.952999	0.951341	0.949730	0.946881	0.947458	0.947507
R²	0.908157	0.904874	0.901881	0.896313	0.897537	0.897657

Bi-LSTM (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.524607	5.632932	5.771882	5.954062	6.055177	6.096319
RMSE	9.129195	9.285857	9.423597	9.640355	9.636107	9.591586
MAPE	0.132028	0.140429	0.127086	0.122740	0.127745	0.141035
R	0.953077	0.951397	0.949857	0.946916	0.947363	0.947457
R²	0.908075	0.904848	0.902081	0.896346	0.897331	0.897549

FI-RNNs (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.520539	5.626097	5.768340	5.932466	6.028442	6.048296
RMSE	9.116093	9.267610	9.432268	9.622239	9.603523	9.569430
MAPE	0.137907	0.144544	0.130363	0.123199	0.129686	0.140039
R	0.953136	0.951567	0.949828	0.947121	0.947781	0.947735
R²	0.908339	0.905221	0.901901	0.896736	0.898025	0.898022

PSPNN (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.386255	5.468801	5.567571	5.716054	5.768782	5.749586
RMSE	8.880815	9.022358	9.145341	9.307195	9.271415	9.167632
MAPE	0.130217	0.135699	0.120082	0.114041	0.116743	0.128589
R	0.955524	0.954207	0.952829	0.950642	0.951668	0.952150
R²	0.913009	0.910171	0.907778	0.903387	0.904956	0.906406

MDL (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.429035	5.509909	5.627575	5.784230	5.813673	5.804645
RMSE	8.916788	9.076988	9.176666	9.381263	9.304234	9.256170
MAPE	0.130555	0.134097	0.120055	0.115990	0.117240	0.129795
R	0.955434	0.953698	0.952556	0.949963	0.951238	0.951537
R²	0.912303	0.909080	0.907145	0.901843	0.904282	0.904590

T-GCN (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.670292	5.675068	5.746703	5.863256	5.888195	5.831183
RMSE	9.134161	9.173527	9.245162	9.387154	9.317363	9.195807
MAPE	0.138232	0.141509	0.124293	0.118078	0.119514	0.132048
R	0.953061	0.952621	0.951832	0.949719	0.950864	0.951782
R²	0.907975	0.907136	0.905754	0.901720	0.904011	0.905830

DCRNN (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.808297	5.788443	5.794671	5.819892	5.836586	5.867232
RMSE	9.389112	9.371624	9.375622	9.404970	9.420036	9.460167
MAPE	0.137371	0.133297	0.133757	0.134122	0.133960	0.135183
R	0.950188	0.950462	0.950404	0.950128	0.949942	0.949522
R²	0.902386	0.902738	0.902651	0.902037	0.901720	0.900876

AST-GAT (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.248924	5.258951	5.329265	5.428566	5.421416	5.314834
RMSE	8.705578	8.784430	8.828416	8.943468	8.809244	8.712925
MAPE	0.124102	0.128212	0.114900	0.110572	0.110653	0.121733
R	0.957302	0.956514	0.956069	0.954439	0.956152	0.956983
R²	0.916408	0.914847	0.914059	0.910791	0.914195	0.915460

GMAN (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.233438	5.208744	5.256062	5.329036	5.347667	5.204790
RMSE	8.730351	8.795372	8.797104	8.908348	8.818618	8.598270
MAPE	0.130036	0.131432	0.116220	0.107743	0.111887	0.117915
R	0.957150	0.956506	0.956505	0.954913	0.956238	0.958076
R²	0.915932	0.914634	0.914668	0.911490	0.914013	0.917670

GMAN_1 (without decoder on spatiotemporal attention)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.234585	5.178617	5.211098	5.287373	5.322027	5.223792
RMSE	8.734820	8.744925	8.779705	8.879185	8.817018	8.640688
MAPE	0.130630	0.128092	0.114170	0.106051	0.110179	0.120951
R	0.957045	0.956970	0.956621	0.955153	0.956184	0.957628
R²	0.915846	0.915611	0.915005	0.912069	0.914044	0.916856

ST-GRAT (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.126567	5.224280	5.335337	5.456082	5.562169	5.662242
RMSE	8.719842	8.848616	8.980233	9.120206	9.243639	9.359533
MAPE	0.117111	0.119548	9.243639	0.124147	0.126001	0.127665
R	0.957088	0.956256	0.126001	0.954598	0.953828	0.953095
R²	0.915784	0.913271	0.910662	0.907852	0.905335	0.902939

STGIN (Multi-steps)

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.175827	5.100160	5.149714	5.222380	5.245451	5.150738
RMSE	8.644908	8.640150	8.701153	8.819650	8.724352	8.527315
MAPE	0.126175	0.127834	0.112100	0.104818	0.108805	0.117945
R	0.957921	0.957968	0.957378	0.955765	0.957101	0.958707
R²	0.917550	0.917624	0.916518	0.913230	0.915819	0.919006

STGIN_1 (Multi-steps) without semantic transformation

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.177275	5.109780	5.187932	5.239958	5.257546	5.156742
RMSE	8.631874	8.640737	8.729155	8.806985	8.720966	8.515376
MAPE	0.132412	0.131180	0.117357	0.109338	0.113217	0.120528
R	0.958046	0.957964	0.957082	0.955788	0.957056	0.958787
R²	0.917799	0.917613	0.915979	0.913479	0.915884	0.919233

STGIN_2 (Multi-steps) uses an encoder of GMAN to replace the of ST-Block STGIN

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.195487	5.116568	5.174414	5.246917	5.274146	5.178238
RMSE	8.718395	8.693000	8.744741	8.864868	8.775015	8.614857
MAPE	0.127035	0.129896	0.113813	0.105365	0.108400	0.119232
R	0.957350	0.957625	0.957099	0.955481	0.956771	0.958061
R²	0.916143	0.916614	0.915679	0.912338	0.914838	0.917335

STGIN_3 (Multi-steps) dynamic step-by-step method to replace the generative inference of STGIN

评价指标	6-1 steps	6-2 steps	6-3 steps	6-4 steps	6-5 steps	6-6 steps
MAE	5.182895	5.167289	5.197474	5.254837	5.308890	5.164660
RMSE	8.666809	8.727748	8.737943	8.846064	8.761161	8.555215
MAPE	0.126229	0.129717	0.114406	0.107089	0.108753	0.117827
R	0.957731	0.957168	0.957062	0.955487	0.956708	0.958486
R²	0.917151	0.915942	0.915812	0.912723	0.915129	0.918493

代码更正

如果你想保证模型的精准度更高，请将 bridge.py 中的代码，改为以下代码 (block set to 1， and input length set to 12)

新版STGIN映射到STGIN_4权重weights/参数上，以此类推，STGIN_3映射到STGIN_7权重weights/参数上，

      # Blocks
      for i in range(self.num_blocks):
          with tf.variable_scope("num_blocks_{}".format(i)):
              # Multihead Attention
              X_Q = multihead_attention(queries=X_Q, # future time steps
                                      keys=X_P,    # historical time steps
                                      values= X,   # historical inputs
                                      num_units=self.hidden_units,
                                      num_heads= self.num_heads, # self.num_heads
                                      dropout_rate=self.dropout_rate,
                                      is_training=self.is_training)
              # Feed Forward
              X_Q = feedforward(X_Q, num_units=[4 * self.hidden_units, self.hidden_units])
  X = tf.reshape(X_Q,shape=[-1, self.site_num, self.output_length, self.hidden_units])

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
API		API
baseline		baseline
data		data
models		models
results		results
utils		utils
weights		weights
.DS_Store		.DS_Store
.gitignore		.gitignore
README.md		README.md
inference.py		inference.py
requirements.txt		requirements.txt
run_baseline.py		run_baseline.py
run_dynamic_inference.py		run_dynamic_inference.py
run_gman.py		run_gman.py
run_train.py		run_train.py
test.py		test.py
visualize.py		visualize.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Traffic-speed-prediction

注意事项

Experimental Results

HA (Multi-steps)

ARIMA (Multi-steps)

SVM (Multi-steps)

LSTM (Multi-steps)

Bi-LSTM (Multi-steps)

FI-RNNs (Multi-steps)

PSPNN (Multi-steps)

MDL (Multi-steps)

T-GCN (Multi-steps)

DCRNN (Multi-steps)

AST-GAT (Multi-steps)

GMAN (Multi-steps)

GMAN_1 (without decoder on spatiotemporal attention)

ST-GRAT (Multi-steps)

STGIN (Multi-steps)

STGIN_1 (Multi-steps) without semantic transformation

STGIN_2 (Multi-steps) uses an encoder of GMAN to replace the of ST-Block STGIN

STGIN_3 (Multi-steps) dynamic step-by-step method to replace the generative inference of STGIN

代码更正

如果你想保证模型的精准度更高，请将 bridge.py 中的代码，改为以下代码 (block set to 1， and input length set to 12)

About

Releases

Packages

Languages

bjj1998/STGIN

Folders and files

Latest commit

History

Repository files navigation

Traffic-speed-prediction

注意事项

Experimental Results

HA (Multi-steps)

ARIMA (Multi-steps)

SVM (Multi-steps)

LSTM (Multi-steps)

Bi-LSTM (Multi-steps)

FI-RNNs (Multi-steps)

PSPNN (Multi-steps)

MDL (Multi-steps)

T-GCN (Multi-steps)

DCRNN (Multi-steps)

AST-GAT (Multi-steps)

GMAN (Multi-steps)

GMAN_1 (without decoder on spatiotemporal attention)

ST-GRAT (Multi-steps)

STGIN (Multi-steps)

STGIN_1 (Multi-steps) without semantic transformation

STGIN_2 (Multi-steps) uses an encoder of GMAN to replace the of ST-Block STGIN

STGIN_3 (Multi-steps) dynamic step-by-step method to replace the generative inference of STGIN

代码更正

如果你想保证模型的精准度更高，请将 bridge.py 中的代码，改为以下代码 (block set to 1， and input length set to 12)

About

Resources

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages