关键词: 果蔬生鲜物流指数, 物流需求预测, 支持向量机, 皮尔逊交叉法, 蝙蝠算法

Abstract: Changes in logistics demand information in the fresh fruit and vegetable market belong to a very complex nonlinear process, with numerous influencing variables that are difficult to quantify. This article attempts to construct the Fruit & Vegetable Logistics Index (FVLI), which is mainly used to reflect demand level of local fruit and vegetable market and is also an important indicator to reflect changes in regional logistics market information. In theory, this study can predict demand for logistics and improve overall efficiency of the fruit and vegetable fresh supply chain. In practice, it can avoid bullwhip effect caused by imbalance of supply and demand, which can lead to price spikes in fruit and vegetable fresh markets, providing reference for government and business decision-makers.
The first step of research process is to use Internet big data search technology to select influencing factors through online search, find six major influencing factors such as logistics, economy, supply, demand, policies and regulations, and level of scientific and technological development, and then build a network keyword lexicon related to the index. 36 secondary indicators are obtained from six major primary indicators. The second step is to avoid multicollinearity caused by high correlation between variables that cannot pass the significance test. Pearson correlation analysis and stepwise regression are used to select the final predictor for correlation between variables, and significance levels can be used. The third step is to combine penalty coefficient, insensitivity, and kernel parameters determined by bat algorithm, normalize crawler data, and construct an evaluation system for measuring prediction accuracy. The fourth step is to evaluate the fitness of parameters determined by bat algorithm, compare results before and after optimization. The fifth step is to use data for eight years as actual values to obtain estimated predicted values. The results are then reversed to obtain test result data, which preliminarily indicates that theBA-SVR hybrid prediction model has strong robustness, fast convergence speed, and high prediction accuracy. Finally, BA-SVR hybrid model will be used as benchmark model for comparison with traditional models and neural networks.
The article attempts to construct a demand index for fresh fruit and vegetable logistics, and combines machine learning and statistical knowledge to provide an improved method that can be used for predicting demand for fresh fruit and vegetable logistics. The optimization model has good generalization abilities, among which innovation is mainly reflected in the following three points: Firstly, to construct a prediction index for the demand for fresh fruits and vegetables, and propose applicable scope and assumed conditions. The second point is to combine network big data and Python software for data crawler collection and processing. The third point is to use classical machine learning method of support vector machine to optimize free parameters that exist in support vector regression using the bat algorithm. By updating and iterating, the optimization value of free parameters is finally determined, and a BA-SVR hybrid prediction model is constructed. By taking advantage of Bat Algorithm (BA) in automatically updating iterative parameters, it is introduced into the Support Vector Regression model to optimize the free parameter values in the SVR model, simulate and empirically predict demand change trend of fresh fruit and vegetable in Beijing, which has a good theoretical value and practical application significance.

Key words: fresh fruit and vegetable logistics index, logistics demand forecast, support vector machine, Pearson cross method, bat algorithm