基于粒子群优化的小波神经网络的停车位数量短时预测

APPLICATION TECHNOLOGY OF PARKING PLACE NUMBERS SHORT-TERM PREDICTION BASED ON WNN OPTIMISED BY PSO AND ITS STUDY

针对停车场有效停车位的短时间预测精度低的问题,首先提出基于梯度下降法的小波神经网络模型,并且用粒子群优化算法对小波神经网络的参数作进一步的优化。用天津站后广场地下停车场的历史数据进行实验,结果表明该模型能对短时间内有效停车位数进行较准确的预测,且用粒子群优化算法对小波神经网络的参数优化后预测的平均绝对误差减小了5.23,平均相对误差减小了2.11%,最大相对误差降低了10.39%。实验结果表明,该模型能较准确地预测短时间内停车位数量,且优化后预测精度得到了进一步的提高。

Aiming at the problem of low accuracy in short-term prediction on effective parking places in parking lot, first we proposed the gradient descent method-based wavelet neural network model, and used particle swarm optimisation to further optimise the parameters of WNN. The historical data of underground parking lot in Tianjin station rear plaza was used in the experiment, results showed that the proposed model was able to make rather,precise prediction on effective parking places in short term. Moreover, after optimising wavelet neural network with particle swarm optimisation, the average absolute error of prediction decreased 5.23, the average relative error decreased 2.11% , and the maximum relative error decreased 10.39%. Experimental result demonstrated that the model can more accurately predict the parking places in short term, and the optimised prediction accuracy gained further improvement.