Study On Pharmaceutical Forecasting Model Based On Computational Intelligence And Its Applications In Pharmaceutical Intelligent Logistics Platform

With rapid development of Internet and information technology,pharmaceutical industry is merging with Internet in progress.Internet、Big Data and Artificial Intelligence,are firmly ongoing change the operation modes of pharmaceutical research、circulation、product tracing、managerial supervision and feedback information.This thesis based on the Intelligent Logistics Public Information Platform for Medicine Circulating which is the first national pilot project in Chongqing carrying forward the supply chain collaboration and controlling amount of drugs、medical apparatus and instruments from the original production to terminal market.This platform implements interconnection of medical supply chain and information sharing based on standard unified coding of medical apparatus and instruments、E-commerce、logistics distribution、and online settlement.Via continuous operation for 5 years,it has accumulated mass of medical instruments traction data,how to make use of those data and expand platform service are the most important research topic.Therefore,this paper will focus on using intelligent algorithm models to successfully forecast drug sales that realize the following extended application services,such as key species of medication warning,pharmaceutical inventory warning,flow direction warning,flexible production decision-making functions acting on pharmaceutical enterprises、medical institutions and regulatory institutions.This thesis firstly analyses in aspects of thought export、implementation steps and applications to figure out basic comprehension of traditional pharmaceutical sales forecasting models,in order to accomplish researching and establishing new pharmaceutical sales forecasting ways based on computational intelligence.From the point of analysis results,pharmaceutical sales data is easily influenced by many factors,for instance policies、health emergencies、changing seasons,that results in disappointed forecasting consequences.With purpose of solving these matters,this thesis proposes two improved forecasting models named IPPFA-SVM and CSMFPA-SVM according to historical transaction data.These models have the following novel characteristics:(1)Based on Inverse Proportion Polynomials Firefly Algorithm-Support Vector Machine(IPPFA-SVM),solves inaccurate output values and monotone decreasing in exponential rate of traditional model caused by limited distance between light and attraction intensity.For overcome this shortness,this thesis uses inverse polynomial function to improve Firefly Algorithm,combine with SVM,get a new algorithm.In the process of research,the largest sale volume and widest usage of pharmaceutical transaction data is chosen as experimental data.By comparison of traditional SVM、BP、Polynomial regression、ARIMA and Hybrid model,it is proved that IPPFA-SVM shows positive results in drug sale forecasting and even low error rate.(2)On the basis of analyzing incorporation between Flower Pollination Algorithm and Support Vector Machine,Chaotic Sine Map Function Algorithm is introduced to optimize low accuracy and local minimum and generate Chaotic Sine Map Flower Pollination Algorithm-Support Vector Machine model(IPPFA-SVM).By selecting medical transaction data of largest sale volume and widest usage aspect as experimental data,this thesis conducts tests and verifies according week、month、season and year.After that,in comparison with single SVM、BP、Polynomial regression、ARIMA and Hybrid model,CSMFPA-SVM reveals a better results in drug sale forecasting with less than 10% error rate as well as a even better than IPPFA-SVM.(3)We combine IPPFA-SVM and CSMFPA-SVM models with platform functions and apply them in key species of medication warning,pharmaceutical inventory warning,flow direction warning,flexible production decision-making.In the field of key species of medication warning,on the basis of IPPFA-SVM,it makes comparison of actual and forecasting volume.While deviation value is more than 20%,it would release warning information to medical institutions and regulatory institutions that effective controls substance abuse and rationally utilization of medication.For pharmaceutical inventory warning,based on CSMFPA-SVM to predict all catalog pharmaceutical sales for single medical institute,this application customize a upper and lower prediction bound with a floating ratio,then automatically generate inventory report to realize order supplement of residual quantity and meticulous management.Furthermore,we use the both forecasting models in pharmaceutical flow direction prediction,integrate the whole Chongqing production of medical instruments with distribution data.Therefore,it brings accurate results in terminal flow direction forecasting for manufacturing and operating enterprises;finally,these two forecasting models are used to handle flexible production decision making.Based on historical transaction data,it accomplishes to make predictions of needed drug production quantities and dispatch related production factors.After considering others factors,such as policies、season、climate,we propose two new forecasting models,IPPFA-SVM and CSMFPA-SVM based on pharmacological categories,drug attributes and feature.Those two models contain better stability and reliability than SVM、BP、Polynomial regression、ARIMA and Hybrid model.The performance are acceptable in fields of key species of medication warning,pharmaceutical inventory warning,flow direction warning and flexible production decision making.It is absolute that with further progress of these two models in pharmaceuticals industry,they would display high value on information synergism application and improve present operation mode on research,development,sales market,tracing,supervision and control,feedback method.