基于VMD-LSTM神经网络的有效停车泊位预测
A Method for Forecasting Available Parking Spaces Based on VMD-LSTM Neural Network

有效停车泊位预测是停车诱导系统的重要内容，同时也是智能交通领域的重要问题之一，然而当前大多数有效停车泊位预测模型针对停车泊位数据的非线性和非平稳性考虑不足，预测误差较大，预测精度较差。为了提高有效停车泊位预测精度，针对有效停车泊位的变化特征，提出了一种基于变分模态分解–长短时记忆网络(VMD-LSTM)的组合预测模型。以路边停车位历史有效泊位数据为基础，先将有效停车泊位数据通过变分模态分解成若干个线性的、平稳的模态分量，将各个分量分别划分为训练集和测试集，然后使用长短时记忆网络分别预测各分量，再对预测结果进行叠加、重构，得到最终预测结果。为验证模型的有效性，选取实际路边停车位数据作为研究对象进行了实例验证。结果表明，所提出的变分模态分解–长短时记忆网络组合预测模型的R2、平均绝对误差(MAE)、均方根误差(RMSE)、平均绝对百分比误差(MAPE)等指标均低于其他预测模型，预测精度最高，预测误差最小。研究表明，变分模态分解方法可有效降低停车泊位数据的非线性和非平稳性，与不使用模态分解方法的神经网络预测模型相比，预测精度有40%~62%的提高，精确度明显提高，变分模态分解–长短时记忆网络组合预测模型能很好地反映有效停车泊位变化的规律和趋势。
Available Parking Spaces prediction constitutes a crucial component of parking guidance systems and represents one of the significant issues in the field of intelligent transportation. However, most current prediction models for available parking spaces inadequately address the nonlinearity and non-stationarity of parking spaces data, leading to substantial prediction errors and poor prediction accuracy. To enhance the prediction accuracy of available parking spaces, a hybrid prediction model based on Variational Mode Decomposition (VMD) and Long Short-Term Memory (LSTM) networks is proposed, tailored to the changing characteristics of available parking spaces. Using historical available parking spaces data from curbside parking as the foundation, the available parking spaces data is first decomposed into several linear and stationary modal components through VMD. Each component is then divided into training and testing sets, and LSTM networks are employed to predict each component separately. The prediction results are subsequently aggregated and reconstructed to obtain the final prediction. To validate the effectiveness of the model, actual curbside parking spaces data is selected for case verification. The results indicate that the proposed VMD-LSTM hybrid prediction model outperforms other prediction models in terms of metrics such as R2, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE), achieving the highest prediction accuracy and the smallest prediction error. The research reveals that the VMD method can effectively reduce the nonlinearity and non-stationarity of parking spaces data. Compared with neural network prediction models without modal decomposition, the prediction accuracy is improved by 40%~62%, demonstrating a significant increase in precision. The VMD-LSTM hybrid prediction model successfully captures the