| The development of unmanned storage technology in China is very rapid,and the application scale is expanding rapidly.The most popular unmanned scheme is the cargo to person mode with AGV(automated guided vehicle)as the core.The AGV brands adopted by enterprises are also different,but the structure of AGV is basically the same.In terms of power supply module,rechargeable lithium battery is chosen as the solution.We have conducted in-depth and cooperative discussion and research with one of the companies that provides unmanned warehouse solutions.It is found that the aging problem of AGV battery is an important factor affecting the reliability of AGV and unmanned warehouse system The battery life parameters provided by the company are not optimized specifically for the AGV usage scenario,but are only a general standard parameter.As the battery life is greatly affected by the specific use strategy,violent adoption of the default age parameter provided by the manufacturer can not accurately measure the actual aging degree of the battery,and there will be errors in the measurement of the aging degree of the battery A huge waste of cost.From the perspective of data-driven,this paper puts forward a method to complete the missing age parameters of AGV battery,and a method to diagnose whether the AGV battery is aging through the relevant data when AGV is working.It is hoped that this method can relatively accurately determine the age of AGV and the aging process of battery,save costs for enterprises and ensure the stability of AGV system at the same time.After determining the two major issues in this paper,we systematically review the literature on the two problems from the ideas of Feature Engineering,the methods,algorithms and models used in the same field or related fields in recent years,and carefully evaluate the advantages and disadvantages of the work done in these literatures.Through reviewing the existing literature,the following three conclusions are summarized:(1)at present,there are few researches on aging diagnosis of data-driven lithium batteries;(2)in the existing literature,few data-driven diagnosis of lithium battery aging is carried out combining with the data of specific business scenarios;(3)LightGBM algorithm is very popular in the current machine learning algorithm competition,but data-driven lithium battery is very popular This algorithm is rarely used in the literature of aging diagnosis.Then,in cooperation with a large unmanned warehouse solution provider.we obtained part of the working data of AGV and the corresponding battery data.On this basis,we made the following work in view of the urgent problems of the two enterprises:(1)In the clustering experiment of battery brands,based on the current data distribution of different brands of batteries in different states under charging and discharging.the feature engineering scheme of extracting extreme value.quantile and average value of battery current value under different charging and discharging states is extracted.After applying Kmeans clustering algorithm,the clustering effect is better,and different brands of batteries can be more accurately clustered In order to further improve the accuracy of the model.separate the battery models of different brands in the following experiments.(2)In the experiment of battery full charge discharge times.through the decomposition analysis of battery discharge state,the statistical characteristics of battery discharge time under different discharge states and some key information(such as current,temperature.speed,shelf weight)in the corresponding discharge process are extracted and calculated.The feature engineering scheme can accurately calculate the full discharge times of the battery In the application of LightGBM as training prediction model.the performance in accuracy and running time is better,which can meet the needs of enterprise production.(3)In order to solve the problem of AGV battery aging monitoring mechanism which is more specific for unmanned warehouse environment.we introduce the number of full and full discharge of each battery on the basis of the time length and related statistical characteristics of seven discharge states proposed in the previous experiment.Through the model.we learn the relationship between these characteristics and the percentage of power consumption decline,and complete the training of classification model Through the introduction of oversampling sample enhancement technology to solve the problem of sample imbalance,the accuracy of the model is further improved.The result of the final classification model is obviously better than the battery aging judgment mechanism based on full discharge times provided by battery suppliers in F1 score.which provides enterprises with a reference battery aging determination mechanism based on operation data.The main conclusions of this work include:(1)We focus on the current of different brands of batteries in different working states to carry out feature engineering.and use k-means to carry out unsupervised classification.which has achieved good results.In the promotion process of industry 4.0,it provides an idea to solve the problem of data loss in many traditional enterprises,especially in the scenario where lithium batteries are installed in the main equipment.(2)Aiming at the problem that the missing value of the full and full discharge times of the battery is supplemented,the solution of using lightgbm algorithm to segment different working states and then carry out feature engineering is proposed,which has higher accuracy and faster calculation speed.It provides a feasible and effective solution for the improvement of AGV battery life monitoring system based on AGV unmanned storage system The case and Research Foundation provide a new and reliable model and research ideas for the industrial manufacturing enterprises in the improvement of the life monitoring system of rechargeable lithium batteries.At the same time,it will also provide an important feature for our next intelligent diagnosis algorithm of battery aging.(3)Based on the existing monitoring mechanism of battery aging,a new method of battery aging diagnosis is proposed based on AGV operation data and battery related information data.Some new attempts are made in feature construction and algorithm application.The number of full fill discharges in the previous chapter is added to the model as an important feature.The oversampling+lightgbm algorithm is proposed According to the characteristic engineering logic proposed by us,the accuracy of the proposed feature engineering logic is obviously higher than that of the original judgment rules,and it is also significantly better than other machine learning algorithms under the same feature engineering logic.It provides a feasible battery aging monitoring scheme for the unmanned storage enterprises based on AGV,and provides a new reference for all enterprises with high battery replacement cost in the direction of battery aging monitoring Ideas to further save battery costs for enterprises. |
PDF Full Text Request