The invention discloses a training method of a road condition prediction model and a road condition prediction method and device, and the method comprises the steps: carrying out the data transformation of a historical road condition training set, obtaining a plurality of target image data, carrying out the feature extraction of the plurality of target image data through a preset residual image convolution model, obtaining a plurality of residual images, and carrying out the feature extraction of the plurality of target image data. Capturing data characteristics with direction relevance by utilizing a spatial dependency relationship of the graph data so as to combine time relevance characteristics of historical data of a plurality of time periods; and performing fusion and full connection on the plurality of residual images, outputting a visual road condition prediction value, calculating a loss function of the residual image convolution model according to the visual road condition prediction value, and updating model parameters of the residual image convolution model based on the loss function until the residual image convolution model reaches a preset convergence condition, thereby obtaining a road condition prediction model. Therefore, global features such as time and space are fully utilized, a prediction model with visual road condition features is constructed, model prediction results are enriched, and model prediction accuracy is improved.
本申请公开了一种路况预测模型的训练方法、路况预测方法及装置,通过对历史路况训练集进行数据变换,得到多个目标图数据,利用预设的残差图卷积模型,对多个目标图数据进行特征提取,得到多个残差图,从而利用图数据的空间依赖关系捕捉具有方向关联性的数据特征,以结合多个时段的历史数据所具有的时间关联性特征;对多个残差图进行融合和全连接,输出视觉路况预测值,并根据视觉路况预测值,计算残差图卷积模型的损失函数,以及基于损失函数更新残差图卷积模型的模型参数,直至残差图卷积模型达到预设收敛条件,得到路况预测模型,从而充分利用时间和空间等全局特征,构建具备视觉路况特征的预测模型,丰富模型预测结果和提高模型预测准确度。
Training method of road condition prediction model, and road condition prediction method and device
路况预测模型的训练方法、路况预测方法及装置
2022-04-05
Patent
Electronic Resource
Chinese
European Patent Office | 2021
|Traffic road condition prediction method, classification model training method and medium
European Patent Office | 2024
|