关键词: 模式识别, 交通流短时预测, K-近邻搜索, 距离权重, 均方根误差

Abstract: In the short-term traffic flow forecasting based on the weighted traffic pattern recognition algorithm(WPRA)，the weights of different historical state values are distinguished according to the time-interval characteristics of historical traffic patterns，but in practical application，the subjective setting of the weight values reduces the reliability of this method． In this paper，by analyzing the core principle of the data-driven non-parametric-regression traffic flow forecasting algorithm and by improving the forecasting algorithm aiming at the uncertainty of the time-interval-characteristic weights of WPRA，a distance-based weighted pattern recognition algorithm (DWPRA) is proposed to forecast the short-term traffic flow． Finally，the root mean square error between real traffic flows and predicted ones is introduced to verify the proposed algorithm． The results show that，at the same neighbor number K，the root mean square error of DWPRA is 4.8% ～7.1% lower than that of WPRA，which proves the effectiveness of DWPRA．

Key words: pattern recognition, short-term traffic flow forecasting, K-nearest neighbor search, distance weight, root mean square error