Express delivery volume prediction method, device, equipment and storage medium

Abstract

The present invention relates to the technical field of logistics prediction, and discloses a method, device, equipment and storage medium for predicting the quantity of express parcels. By performing differential calculation on the acquired historical data, a plurality of differential values ​​are obtained, and the historical data is formed into sample data based on the differential values. The model is trained according to the AR model algorithm, the differential values ​​and the sample data to generate a quantity prediction model. After determining the cycle time to be predicted, the corresponding cycle parameters and the historical target data set are calculated, and the quantity prediction model is used to perform prediction to obtain the predicted quantity of parcels for the cycle time. The use of differentials to generate samples improves the stability of the samples, so that the accuracy of the obtained model is higher, and the accuracy of the predicted quantity of parcels is improved. At the same time, through AR model prediction, manual operations are reduced, and work efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of logistics forecasting technology, and in particular to a method, device, equipment and storage medium for predicting the quantity of express parcels. Background Art

[0002] With the development of science and technology and the improvement of living standards, online shopping has become a lifestyle trend for users, which requires merchants to process orders and transport them in a timely manner. At present, in order to ensure the real-time delivery of orders, merchants will choose to entrust corresponding logistics companies to achieve this, thereby ensuring the transportation of orders.

[0003] However, the current order delivery and transportation of logistics companies are mainly arranged according to the actual order volume. At the same time, they also adopt an order volume prediction scheme to facilitate the subsequent transportation and delivery arrangements. However, in the prior art, predictions and pre-arrangements are mainly made based on holidays or merchant promotions. This prediction method is too simple, and its prediction can only be calculated according to fixed set rules, and cannot be dynamically adjusted according to actual conditions, resulting in a large difference between the predicted order volume and the actual one, and the prediction accuracy is not high. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem that the prediction results of the existing piece quantity prediction scheme are of low accuracy.

[0005] A first aspect of the present invention provides an express delivery volume prediction method based on an AR model, which is applied to an express delivery management platform. The express delivery volume prediction method includes: reading historical data on the express delivery volume from a background database in the express delivery management platform, and extracting target data sets of at least two historical cycles from the historical data; performing differential processing on at least two of the target data sets using a differential method to obtain multiple differential values, and optimizing at least two of the target data sets based on the differential values ​​to obtain a sample data set; performing fitting training and learning on the multiple differential values ​​and the sample data set using a preset AR model algorithm to obtain a volume prediction model; determining a cycle time to be predicted, and calculating corresponding cycle parameters and historical target data sets based on the cycle time; and using the volume prediction model to predict the express delivery volume within the cycle time according to the cycle parameters and the historical target data set to obtain a predicted volume.

[0006] Optionally, in a first implementation manner of the first aspect of the present invention, extracting target data sets of at least two historical periods from the historical data includes: extracting waybills from the historical data, scanning and identifying the barcodes in each waybill, and determining the type of each waybill and the processing time of the waybill, wherein the types include receiving waybills and sending waybills; classifying the historical data according to the types to obtain a historical receiving data set and a historical sending data set; identifying abnormal data and redundant information for the historical receiving data set and the historical sending data set according to a preset data filtering rule, and marking corresponding identification information to obtain a corresponding identification data set; correcting corresponding data or information from the corresponding identification data set according to the identification information, and filling blank data in the corrected data set by replacing adjacent data to obtain a new data set; identifying the order number information and address information in the new data set, and deleting them to obtain a first historical receiving data set and a first historical sending data set; merging and dividing the first historical receiving data set and the first historical sending data set according to the processing time by daily periods to obtain target data sets of at least two historical periods, wherein the target data set includes at least one waybill data.

[0007] Optionally, in a second implementation method of the first aspect of the present invention, the historical receiving data set and the historical sending data set are merged and divided according to the processing time on a daily basis to obtain target data sets of at least two historical periods, including: according to the processing time, the waybill data in the first historical receiving data set and the first historical sending data set are sorted in chronological order to obtain a historical receiving data sequence and a historical sending data sequence; the waybill data in the historical receiving data sequence and the historical sending data sequence with the same processing time are merged on a daily basis to form at least two small data sets, wherein the small data set contains at least one of a receiving waybill and a sending waybill; the total quantity of the receiving waybill and / or the sending waybill in each small data set is counted, and the quantity is marked to generate at least two target data sets.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the differential processing of at least two of the target data sets is performed using the differential method to obtain multiple differential values, and the at least two target data sets are optimized based on the differential values ​​to obtain a sample data set, including: performing multiple logarithmic calculations on the total piece quantity of each target data set to obtain multiple logarithmic values ​​of each target data set; performing differential calculations on the multiple logarithmic values ​​using the differential method to obtain multiple differential values, wherein the differential calculation is to subtract the same logarithmic values ​​of the total piece quantity of two consecutive target data sets; judging whether the multiple differential values ​​corresponding to each target data set meet a preset threshold; if not, eliminating the target data sets whose differential values ​​do not meet the threshold to obtain a sample data set.

[0009] Optionally, in a fourth implementation method of the first aspect of the present invention, the differential value includes a first-order differential value and a second-order differential value, and the use of a preset AR model algorithm to perform fitting training and learning on multiple differential values ​​and the sample data set to obtain a piece quantity prediction model includes: performing self-regression calculation on the waybill data in the sample data set to obtain corresponding regression variables; performing random linear combinations of the waybill data according to each regression variable, and performing regression analysis on the linear combinations to obtain autocorrelation coefficients between each waybill data; outputting the sample data set to a function formula corresponding to a preset AR model algorithm, and performing sample fitting training based on the autocorrelation coefficient and the second-order differential value in the differential value to obtain model parameters; and adjusting the function formula based on the model parameters to obtain a piece quantity prediction model.

[0010] Optionally, in a fifth implementation manner of the first aspect of the present invention, after using the preset AR model algorithm to perform fitting training and learning on the multiple differential values ​​and the sample data sets to obtain the piece quantity prediction model, it also includes: selecting at least two waybill data from the sample data set to form a test data set; inputting the test data set into the piece quantity prediction model to predict the piece quantity and obtain a predicted value; calculating the error between the predicted value and the waybill data in the test data set; determining whether the error is within a preset error range; if not, filtering the waybill data whose error is not within the error direction to form a new sample data set; performing secondary training on the piece quantity prediction model based on the new sample data, and outputting the optimal piece quantity prediction model.

[0011] Optionally, in a sixth implementation method of the first aspect of the present invention, the calculation of the corresponding cycle parameters and historical target data set based on the cycle time includes: filtering out waybill data whose processing time is close to the cycle time from at least two of the target data sets to form a historical target data set; calculating the correlation coefficient between each waybill data and the processing time in the historical target data set; determining the autocorrelation coefficient of each waybill data in the historical target data set based on the correlation coefficient to obtain the cycle parameter, wherein the cycle parameter is a coefficient sequence after the autocorrelation coefficients are sorted.

[0012] The second aspect of the present invention provides an AR model-based express delivery volume prediction device, which includes: a collection module, which is used to read the historical data of the express delivery volume from the background database in the express delivery and receiving management platform, and extract at least two historical period target data sets from the historical data; a difference module, which is used to use the difference method to perform difference processing on at least two of the target data sets to obtain multiple difference values, and optimize at least two of the target data sets based on the difference values ​​to obtain a sample data set; a model generation module, which is used to use a preset AR model algorithm to perform fitting training and learning on the multiple obtained difference values ​​and the sample data set to obtain a volume prediction model; a calculation module, which is used to determine the cycle time to be predicted, and calculate the corresponding cycle parameters and the historical target data set based on the cycle time; a prediction module, which is used to use the volume prediction model to predict the express delivery volume within the cycle time according to the cycle parameters and the historical target data set to obtain a predicted volume.

[0013] Optionally, in a first implementation manner of the second aspect of the present invention, the acquisition module is specifically used to: extract waybills from the historical data, and scan and identify the barcodes in each waybill to determine the type of each waybill and the processing time of the waybill, wherein the types include receiving waybills and sending waybills; classify the historical data according to the types to obtain a historical receiving data set and a historical sending data set; identify abnormal data and redundant information for the historical receiving data set and the historical sending data set according to preset data filtering rules, and mark them with corresponding identification information to obtain a corresponding identification data set; correct the corresponding data or information from the corresponding identification data set according to the identification information, and fill the blank data in the corrected data set by replacing adjacent data to obtain a new data set; identify the order number information and address information in the new data set, and delete them to obtain a first historical receiving data set and a first historical sending data set; merge and divide the first historical receiving data set and the first historical sending data set according to the processing time by daily cycles to obtain target data sets of at least two historical cycles, wherein the target data set includes at least one waybill data.

[0014] Optionally, in a second implementation method of the second aspect of the present invention, the collection module is specifically used to: according to the processing time, sort the waybill data in the first historical receiving data set and the first historical sending data set in chronological order to obtain a historical receiving data sequence and a historical sending data sequence; merge the waybill data with the same processing time in the historical receiving data sequence and the historical sending data sequence according to a daily cycle to form at least two small data sets, wherein the small data set contains at least one of a receiving waybill and a sending waybill; count the total number of receiving waybills and / or sending waybills in each small data set, and mark the number of pieces to generate at least two target data sets.

[0015] Optionally, in a third implementation of the second aspect of the present invention, the differential module includes: a logarithmic unit, used to perform multiple logarithmic calculations on the total piece quantity of each target data set to obtain multiple logarithmic values ​​of each target data set; a differential unit, used to perform differential calculations on multiple logarithmic values ​​using a differential method to obtain multiple differential values, wherein the differential calculation is to subtract the same logarithmic values ​​of the total piece quantity of two consecutive target data sets; a judgment unit, used to judge whether the multiple differential values ​​corresponding to each target data set meet a preset threshold; and a sample generation unit, used to eliminate the target data sets corresponding to the differential values ​​that do not meet the threshold when it is judged that the fluctuation coefficient does not meet the threshold, so as to obtain a sample data set.

[0016] Optionally, in a fourth implementation method of the second aspect of the present invention, the differential value includes a first-order differential value and a second-order differential value, and the model generation module includes: a first calculation unit, used to perform self-regression calculation on the waybill data in the sample data set to obtain corresponding regression variables; a regression unit, used to perform random linear combinations of the waybill data according to each regression variable, and perform regression analysis on the linear combinations to obtain autocorrelation coefficients between each waybill data; a training unit, used to output the sample data set to a function formula corresponding to a preset AR model algorithm, and perform sample fitting training based on the autocorrelation coefficient and the second-order differential value in the differential value to obtain model parameters; a generation unit, used to adjust the function formula based on the model parameters to obtain a piece quantity prediction model.

[0017] Optionally, in a fifth implementation of the second aspect of the present invention, the express quantity prediction device based on the AR model also includes: an error calculation module, which is used to select at least two waybill data from the sample data set to form a test data set; input the test data set into the quantity prediction model to predict the quantity and obtain a predicted value; calculate the error between the predicted value and the waybill data in the test data set; determine whether the error is within a preset error range; if not, filter the waybill data whose error is not within the error position to form a new sample data set; perform secondary training on the quantity prediction model based on the new sample data, and output the optimal quantity prediction model.

[0018] Optionally, in a sixth implementation manner of the second aspect of the present invention, the calculation module includes: a screening unit, used to screen out waybill data whose processing time is close to the cycle time from at least two of the target data sets to form a historical target data set; a calculation unit, used to calculate the correlation coefficient between each waybill data and the processing time in the historical target data set; and a determination unit, used to determine the autocorrelation coefficient of each waybill data in the historical target data set according to the correlation coefficient to obtain a cycle parameter, wherein the cycle parameter is a coefficient sequence after the autocorrelation coefficients are sorted.

[0019] The third aspect of the present invention provides an express delivery volume prediction device based on an AR model, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the express delivery volume prediction device based on the AR model executes the above-mentioned express delivery volume prediction method based on the AR model.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions, and when the computer-readable storage medium is run on a computer, it enables the computer to execute the above-mentioned express delivery volume prediction method based on the AR model.

[0021] In the technical solution of the present invention, a plurality of differential values ​​are obtained by performing differential calculation on the acquired historical data, and the historical data is formed into sample data based on the plurality of differential values. The model is trained on the plurality of differential values ​​and the sample data according to the AR model algorithm to generate a piece quantity prediction model. After determining the cycle time to be predicted, the corresponding cycle parameters and the historical target data set are calculated, and the piece quantity prediction model is used for prediction to obtain the predicted piece quantity of the cycle time. The use of differential generation of samples improves the stability of the samples, makes the obtained model more accurate, and improves the accuracy of the predicted piece quantity. At the same time, through AR model prediction, manual operations are reduced, and work efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1It is a schematic diagram of a first embodiment of a method for predicting the express delivery volume based on an AR model in an embodiment of the present invention;

[0023] Figure 2 Schematic diagram of a second embodiment of a method for predicting express delivery volume based on an AR model in an embodiment of the present invention;

[0024] Figure 3 Schematic diagram of a third embodiment of a method for predicting express delivery volume based on an AR model in an embodiment of the present invention;

[0025] Figure 4 Schematic diagram of a fourth embodiment of a method for predicting express delivery volume based on an AR model in an embodiment of the present invention;

[0026] Figure 5 It is a schematic diagram of an embodiment of a device for predicting the quantity of express items based on an AR model in an embodiment of the present invention;

[0027] Figure 6 It is a schematic diagram of another embodiment of the express delivery quantity prediction device based on the AR model in an embodiment of the present invention;

[0028] Figure 7 It is a schematic diagram of an embodiment of a courier quantity prediction device based on an AR model in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The embodiments of the present invention provide a method, device, equipment and storage medium for predicting express delivery volume based on an AR model. By performing multiple logarithmic calculations on historical data and performing differential processing, sample data with higher stability is obtained, and then the AR model algorithm is used to train the volume prediction model to improve the prediction accuracy.

[0030] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1In the first embodiment of the express delivery volume prediction method based on the AR model in the embodiment of the present invention, the implementation steps of the method are as follows:

[0032] 101, reading historical data of the express delivery volume from a background database in the express delivery and receiving management platform, and extracting target data sets of at least two historical periods from the historical data;

[0033] In this embodiment, the historical data is specifically express delivery volume data, which includes two types: receiving and sending. Specifically, when collecting data, a time range is set according to the demand for sample data, and then the corresponding historical records are queried from the background database based on the time range, and the express delivery volume data recorded in the historical records are extracted to form historical data.

[0034] In actual applications, the platform can also be used to connect to the databases of various express statistics agencies, read the corresponding historical data from the database and store it in the background database, that is, read the data on the quantity of pieces in the logistics industry within a certain period of time published by a certain statistics agency, and the quantity of pieces includes the quantity of received pieces and the quantity of sent pieces. In the database, whether online or offline, information on the quantity of delivered pieces and the quantity of received pieces will be stored. This information may include, but is not limited to: the type and time of the quantity of pieces. The time can be stored on a daily basis or according to the specific time entered into the system. Then read from the background database based on the set time range.

[0035] In this embodiment, after acquiring the historical data, the historical data is also cleaned to ensure that there are no blank fields and abnormal data in subsequent sample data, thereby ensuring the accuracy of model training, wherein the cleaning process specifically includes: cleaning historical data, replacing empty data and abnormal data.

[0036] Among them, cleaning historical data, removing unnecessary information from the acquired historical data and replacing abnormal data. Usually before statistical analysis of data, some irregular data needs to be filtered out to ensure the accuracy of the analysis. Data cleaning is a process of reducing data errors and inconsistencies, mainly detecting and deleting or correcting irregular data.

[0037] 102, performing differential processing on at least two target data sets using a differential method to obtain a plurality of differential values, and optimizing the at least two target data sets based on the differential values ​​to obtain a sample data set;

[0038] In this step, the difference method is specifically used to compare the size of the data set and the data set in the two target data sets, and determine whether the two data sets are the same or average according to the comparison result, so as to determine the corresponding difference value. In practical applications, the difference value of the two data sets can also be obtained by subtracting the total quantity of the two data sets. The multiple difference values ​​are specifically different order values ​​obtained by multi-order difference processing. For example, the total quantity of the two pieces before and after is subtracted to obtain the first order difference value, and the first order difference value before and after is subtracted to obtain the second order difference value. Similarly, the third order is calculated based on the second order, and so on. How many order difference values ​​need to be calculated is selected according to actual needs.

[0039] Furthermore, the data sets in at least two target data sets are screened according to multiple differential values ​​to screen out the data set with the smallest and closest differential value to obtain a new target data set, and then a sample data set is generated according to the new target data set, wherein the sample data set includes a training set, a test set and a validation set.

[0040] 103. Use a preset AR model algorithm to perform fitting training and learning on multiple differential values ​​and sample data sets to obtain a piece quantity prediction model;

[0041] In this step, the AR model algorithm is an algorithm of an autoregressive linear model, which is a statistical method for processing time series. It is a process of using itself as a regression variable, that is, using the linear combination of random variables at several previous moments to describe the linear regression model of random variables at a later moment.

[0042] In this embodiment, the process of training the meta-waybill data in the sample data set using the preset AR model algorithm is as follows:

[0043] First, sort the waybill data in the sample data set in chronological order;

[0044] Then, the sorted sample data set is randomly divided into three parts, and each part contains several continuously sorted waybill data;

[0045] Furthermore, autoregressive analysis is performed on several waybill data sets in each part to obtain the tailing of the autocorrelation coefficient of the p-order autoregressive model of the waybill data in the data set and the p-order truncation of the partial autocorrelation coefficient;

[0046] Finally, the parameters in the AR model algorithm are adjusted based on the autocorrelation coefficient, and a part of the waybill data is selected to fit the adjusted AR model algorithm to obtain the final quantity prediction model.

[0047] 104, determining a cycle time to be predicted, and calculating corresponding cycle parameters and a historical target data set based on the cycle time;

[0048] In this embodiment, the cycle time to be predicted is determined according to customer needs. The cycle time can be understood as a time period, which can be one day, one week, one month, etc. One day is preferably selected here. According to the cycle time, the sample data set is intercepted from the time point adjacent to the cycle time, and the total amount of waybill data for multiple days is intercepted. The total amount of waybill data is classified and sorted by day to obtain the historical target data set.

[0049] Furthermore, the correlation between the data changes of the two days before and after the waybill data of each day in the historical target data set is calculated, and then the autocorrelation coefficient between the waybill data of each day in the historical target data set and the cycle time is determined based on the correlation.

[0050] Finally, based on the autocorrelation coefficient, the amount of waybill data in the historical target data set required for prediction is determined to obtain the cycle parameter.

[0051] For example, it is necessary to predict the volume of express parcels on January 1, 2021. This requires obtaining the volume of express parcels in the time period adjacent to January 1, 2021, for example, the total volume of express parcels per day from December 25, 2020 to December 31, 2020; then determine whether January 1, 2021 is a special day or a holiday, and determine whether the change in the historical volume of express parcels affects the volume at the predicted time based on the determined results, and calculate the degree of influence, and then determine the specific degree of influence between the daily volume of express parcels in the historical volume of express parcels and the predicted time, which is the autocorrelation coefficient, and further calculate the percentage of the daily volume of express parcels in the selected historical volume of express parcels based on the autocorrelation coefficient, and use these percentages as period parameters.

[0052] 105. According to the cycle parameters and the historical target data set, the express delivery volume within the cycle time is predicted using the volume prediction model to obtain the predicted volume.

[0053] In this step, the piece quantity prediction model is configured according to the cycle parameters, and then the data in the historical target data set is classified according to the cycle and input into the configured piece quantity prediction model to calculate the predicted piece quantity corresponding to the cycle time.

[0054] In practical applications, if the percentage of 6-day data in the required historical target data set is 30%, 30%, 20%, 10%, 5%, 3%, and 3%, then the corresponding actual sample sets are calculated according to the percentages, and the sample sets are input as prediction data, and then the quantity of pieces in the corresponding prediction period is calculated according to the cycle parameters and the quantity prediction model.

[0055] Through the implementation of the above method, by calculating the fluctuation coefficient between the waybill data in the historical data, sample data with relatively stable changes are selected, thereby reducing the problem of training errors caused by the unrepresentative sample data in model training. Furthermore, after selecting the sample data, the AR model algorithm is used to train and learn the sample data to obtain a piece quantity prediction model, and the piece quantity is predicted based on the piece quantity prediction model. At the same time, it is also necessary to determine the cycle parameters of the prediction cycle time and the required historical target data set. Selecting the model input data in this way greatly improves the accuracy of the prediction.

[0056] See also Figure 2 The second embodiment of the express delivery volume prediction method based on the AR model in the embodiment of the present invention includes:

[0057] 201, reading historical data of express delivery quantity from the background database in the express delivery and receiving management platform;

[0058] 202, extracting the waybills from the historical data, and scanning and identifying the barcodes in each waybill, to determine the type of each waybill and the processing time of the waybill;

[0059] In this step, the types of waybills include receiving waybills and sending waybills. When obtaining waybills, the order in the historical data is scanned one by one by a code scanning device set in the platform or externally set, and of course, it can also be directly parsed from the records stored in the database.

[0060] During parsing, not only the processing time of the waybill needs to be parsed, but also the type of waybill, such as receiving waybill and sending waybill.

[0061] In practical applications, the extraction can also use OCR technology to identify scanned copies of waybills of historical data.

[0062] 203, classifying the historical data according to the type to obtain a historical receipt data set and a historical outgoing data set;

[0063] 204, merging and dividing the historical received data set and the historical sent data set by day according to the processing time to obtain a target data set of at least two historical periods;

[0064] In this embodiment, the target data set includes at least one waybill data, and the waybill data includes both receipt and delivery data. The merging here refers to merging the receipt waybill and the delivery waybill whose processing time belongs to the same day into one data set to obtain multiple target data sets.

[0065] Furthermore, before merging and dividing the historical received mail data set and the historical sent mail data set by day according to the processing time, the method further includes:

[0066] According to the preset data filtering rules, abnormal data and redundant information are respectively identified for the historical receiving data set and the historical sending data set, and corresponding identification information is marked to obtain a corresponding identification data set;

[0067] Correcting the corresponding data or information from the corresponding identification data set according to the identification information, and filling the blank data in the corrected data set by replacing adjacent data to obtain a new data set;

[0068] The order number information and address information in the new data set are identified and deleted to obtain a first historical received data set and a first historical sent data set.

[0069] Further, according to the processing time, the waybill data in the first historical receiving data set and the first historical sending data set are sorted in chronological order to obtain a historical receiving data sequence and a historical sending data sequence;

[0070] Merge the waybill data in the historical receiving data sequence and the historical sending data sequence with the same processing time on a daily basis to form at least two small data sets, wherein the small data set includes at least one of the receiving waybill and the sending waybill;

[0071] The total quantity of the receiving waybills and / or the sending waybills in each small data set is counted and marked to generate at least two target data sets.

[0072] 205, performing differential processing on at least two target data sets using a differential method to obtain a plurality of differential values, and optimizing the at least two target data sets based on the differential values ​​to obtain a sample data set;

[0073] 206, using a preset AR model algorithm to perform fitting training and learning on multiple differential values ​​and sample data sets to obtain a piece quantity prediction model;

[0074] 207, determining a cycle time to be predicted, and calculating corresponding cycle parameters and a historical target data set based on the cycle time;

[0075] 208. According to the cycle parameters and the historical target data set, the express delivery volume within the cycle time is predicted using the volume prediction model to obtain the predicted volume.

[0076] Through the implementation of the above-mentioned method, by performing differential calculation on the acquired historical data, multiple differential values ​​are obtained, and the historical data is formed into sample data based on the multiple differential values. The model is trained on the multiple differential values ​​and sample data according to the AR model algorithm to generate a piece quantity prediction model. After determining the cycle time to be predicted, the corresponding cycle parameters and historical target data sets are calculated, and the piece quantity prediction model is used for prediction to obtain the predicted piece quantity of the cycle time. This prediction method not only improves the accuracy of the prediction, but also can predict the piece quantity in advance, prepare people and vehicles for physical examination, reduce costs, and reduce losses.

[0077] See also Figure 3 The third embodiment of the express delivery volume prediction method based on the AR model in the embodiment of the present invention includes:

[0078] 301, reading historical data of the express delivery volume from a background database in the express delivery and receiving management platform, and extracting target data sets of at least two historical periods from the historical data;

[0079] 302, performing multiple logarithmic calculations on the total number of pieces of each target data set to obtain multiple logarithmic values ​​of each target data set;

[0080] In this step, when calculating the logarithm, specifically, the logarithm of the total number of pieces in the target data set is calculated. Specifically, when extracting the target data set, the outgoing waybill and the incoming waybill are extracted respectively to form the target data set. The optional implementation steps are as follows:

[0081] Extract the waybills from the historical data, scan and identify the barcodes in each waybill, and determine the type and processing time of each waybill, where the types include receiving waybills and sending waybills;

[0082] Classify the historical data by type to obtain a historical receipt data set and a historical outgoing data set;

[0083] According to the preset data filtering rules, abnormal data and redundant information are respectively identified for the historical receiving data set and the historical sending data set, and corresponding identification information is marked to obtain a corresponding identification data set;

[0084] Correcting the corresponding data or information from the corresponding identification data set according to the identification information, and filling the blank data in the corrected data set by replacing adjacent data to obtain a new data set;

[0085] In practical applications, the correction of historical data can be specifically as follows: determine the specific position of the abnormal data in the identification data set based on the identification information; take the specific position as the symmetry center, and select N waybill data with continuous processing time from the left and right sides of the symmetry center; use the discrete regression algorithm to calculate the mode and median of 2N waybill data; replace the abnormal data at the specific position with the mode or median.

[0086] Identify the order number information and address information in the new data set, and delete them to obtain a first historical receipt data set and a first historical delivery data set;

[0087] Merging and dividing the historical receiving data set and the historical sending data set by day according to the processing time to obtain target data sets of at least two historical periods, wherein the target data set includes at least one waybill data;

[0088] According to the processing time, the waybill data in the first historical receiving data set and the first historical sending data set are sorted in chronological order to obtain a historical receiving data sequence and a historical sending data sequence;

[0089] Merge the waybill data in the historical receiving data sequence and the historical sending data sequence with the same processing time on a daily basis to form at least two small data sets, wherein the small data set includes at least one of the receiving waybill and the sending waybill;

[0090] The total quantity of the receiving waybills and / or the sending waybills in each small data set is counted and marked to generate at least two target data sets.

[0091] 303, performing differential calculation on the multiple logarithmic values ​​using a differential method to obtain multiple differential values;

[0092] The difference calculation is to subtract the same-order logarithmic values ​​of the total number of pieces of two consecutive target data sets;

[0093] 304, determining whether a plurality of differential values ​​corresponding to each target data set meet a preset threshold;

[0094] 305, if not satisfied, the target data set corresponding to the difference value not satisfying the threshold is eliminated to obtain a sample data set;

[0095] 306, performing self-regression calculation on the waybill data in the sample data set to obtain corresponding regression variables;

[0096] 307, according to each regression variable, the waybill data is randomly linearly combined, and the linear combination is subjected to regression analysis to obtain the autocorrelation coefficient between each waybill data;

[0097] 308, outputting the sample data set to a function formula corresponding to a preset AR model algorithm, and performing sample fitting training based on the autocorrelation coefficient and the second-order difference value in the difference value to obtain model parameters;

[0098] In practical applications, the differential value is a multi-order differential value. For example, taking the calculation of the second-order differential value as an example, the total number of pieces X in the collected target data set is firstly logarithmically calculated to obtain the logarithmic value, the first-order differential value is calculated based on the logarithmic value, and the second-order differential value is calculated based on the first-order differential value.

[0099] When calculating the first-order difference value, the logarithm of the current target data set is subtracted from the logarithm of the previous target data set to obtain the first-order difference value of the current target data set; the second-order difference value is the first-order difference value of the current target data set minus the first-order difference value of the previous target data set, and the second-order difference value of the current target data set is calculated by analogy, and the difference values ​​of all target data sets are calculated, as shown in the following table:

[0100] date Number of pieces First-order difference value Second order difference 2017 / 1 / 1 X1_1 2017 / 1 / 2 X1_2 X1_2'=X1_2-X1_1 2017 / 1 / 3 X1_3 X1_3'=X1_3-X1_2 X1_3”=X1_3'-X1_2' 2017 / 1 / 4 X1_4 X1_4'=X1_4-X1_3 X1_4”=X1_4'-X1_3' 2017 / 1 / 5 X1_5 X1_5'=X1_5-X1_4 X1_5”=X1_5'-X1_4' 2017 / 1 / 6 X1_6 X1_6'=X1_6-X1_5 X1_6”=X1_6'-X1_5' 2017 / 1 / 7 X1_7 X1_7'=X1_7-X1_6 X1_7”=X1_7'-X1_6' …… …… …… …… 2017 / 12 / 31 X12_31 X12_31'=X12_31-X12_30 X12_31"=X12_31'-X12_30'

[0101] After the second-order difference value is calculated based on the above method, the second-order difference value and sample data are used to perform fitting training through the AR model algorithm to obtain the AR model parameters.

[0102] 309, adjusting the function based on the model parameters to obtain a piece quantity prediction model;

[0103] In this step, the calculation formula of the piece quantity prediction model is as follows:

[0104]

[0105] Among them, X t is the number of express parcels within time t, c is a constant term, ε t is the random error value, is the autocorrelation coefficient, The closer the value is to the period to be predicted, the larger the value is.

[0106] Furthermore, in order to improve the accuracy of the model, it also includes calculating various types of errors by verifying the prediction and adjusting the prediction model based on the errors. Specifically:

[0107] Select at least two waybill data from the sample data set to form a test data set;

[0108] Input the test data set into the piece quantity prediction model to predict the piece quantity and obtain the predicted value;

[0109] Calculate the error between the predicted value and the waybill data in the test dataset;

[0110] Determine whether the error is within the preset error range;

[0111] If not, filter the waybill data whose errors are not within the error direction to form a new sample data set;

[0112] The quantity prediction model is trained again based on the new sample data to output the optimal quantity prediction model.

[0113] 310, determining a cycle time to be predicted, and calculating corresponding cycle parameters and a historical target data set based on the cycle time;

[0114] In this step, specifically, waybill data whose processing time is close to the cycle time is screened out from at least two target data sets to form a historical target data set;

[0115] Calculate the correlation coefficient between each waybill data and processing time in the historical target data set;

[0116] The autocorrelation coefficient of each waybill data in the historical target data set is determined according to the correlation coefficient to obtain a cycle parameter, wherein the cycle parameter is a coefficient sequence after the autocorrelation coefficients are sorted.

[0117] 311. According to the cycle parameters and the historical target data set, the quantity prediction model is used to predict the express quantity within the cycle time to obtain the predicted quantity.

[0118] In the embodiment of the present invention, steps 310-311 are consistent with steps 104-105 in the first embodiment of the above-mentioned express delivery volume prediction method based on the AR model, and are not described in detail here.

[0119] In practical applications, when predicting the quantity of pieces based on the piece quantity prediction model, the differential value in the historical data is specifically predicted and calculated. The piece quantity prediction model finally outputs the P-order differential value, and then the P-1-order differential value is reversed through the P-order differential value until the 0-order differential value is reversed. The 0-order differential value is the final predicted value of the piece quantity. In this embodiment, the reverse process of the piece quantity prediction value is opposite to the calculation process of calculating the multi-order differential value, which will not be repeated here. The specific details are shown in the following table:

[0120]

[0121] In summary, the method provided in the above embodiment realizes the prediction of express delivery volume. By performing multiple logarithmic calculations on historical data and performing differential processing, sample data with higher stability is obtained, and then the AR model algorithm is used to train the volume prediction model, so that the obtained model is more in line with reality. After determining the cycle time to be predicted, the corresponding cycle parameters and historical target data sets are calculated, and the volume prediction model is used to perform predictions to obtain the predicted volume of the cycle time, thereby further improving the prediction accuracy of the express delivery volume.

[0122] See also Figure 4 The fourth embodiment of the express delivery volume prediction method based on the AR model in the embodiment of the present invention includes:

[0123] 401, extract the quantity from the database and obtain historical data;

[0124] In actual applications, the data source for this volume includes the national volume of parcels collected from 2017 to 2020, a total of several pieces of data.

[0125] In this step, after extracting the output quantity, the step also includes: preprocessing the historical data to select a weekly target data set of at least one historical period.

[0126] In practical applications, historical data on parcel volume refers to the parcel volume data stored in the logistics industry, or it can be the parcel volume data in the logistics industry within a certain period of time published by a statistical agency. The parcel volume includes the number of received parcels and the number of sent parcels. In the database, whether online or offline, information on the number of dispatched and received parcels will be stored. This information may include, but is not limited to: the type and time of the parcel volume. The time can be stored by day or by the specific time entered into the system.

[0127] Preprocess the acquired historical data, including cleaning historical data and replacing empty data and abnormal data.

[0128] Among them, cleaning historical data, removing unnecessary information from the acquired historical data and replacing abnormal data. Usually before statistical analysis of data, some irregular data needs to be filtered out to ensure the accuracy of the analysis. Data cleaning is a process of reducing data errors and inconsistencies, mainly detecting and deleting or correcting irregular data.

[0129] In this embodiment, the prediction is mainly for the quantity of pieces, so the order number information and address information contained in the historical data can be removed. In these historical data, empty data or data with abnormal values ​​(such as non-numerical representation) may appear, and these empty data or data with abnormal values ​​are replaced with their adjacent data.

[0130] Specifically, historical data includes the number of parcels received and / or sent, and the information on the number of parcels received (with orders or without orders) and the number of parcels delivered at each outlet can be retrieved from the database according to different business scenarios. The number of parcels received at a certain outlet will be used as test data below, and the date of the historical data is the number of parcels collected during the national Double Eleven period from 2017 to 2020. After data cleaning, the historical data obtained can be shown in the following table.

[0131] Collection Date Number of items received 2017 / 1 / 1 XXXXXX 2017 / 1 / 2 XXXXXX … XXXXXX 2020 / 12 / 31 XXXXXX

[0132] 402, pre-processing the historical data and centralizing the historical data;

[0133] In this step, the preprocessing is specifically to handle the abnormal data, which includes deletion, substitution (continuous variable mean substitution, discrete variable mode and median substitution), interpolation (regression interpolation, multiple interpolation), and at the same time, the abnormal values ​​can be converted into missing values ​​first, and then the missing values ​​are filled. In practice, the abnormal value processing is generally divided into NA missing values ​​or returned to the company for data repair (data repair is the main method).

[0134] 403, take the logarithm twice and perform difference processing;

[0135] In this step, the difference is obtained by subtracting the logarithmic values ​​of the two days before and after. For example, if the quantity of pieces on two days is 1000 and 5000 respectively, when taking the logarithm, the logarithm of 1000 and 5000 is calculated with base 10 to obtain the value, and then the logarithm of the value is calculated with base 10 again to obtain two results, and then the logarithmic results of the two days are compared to calculate the difference.

[0136] 404, training is performed based on the extracted historical data combined with the AR model to obtain a prediction model;

[0137] In this step, the AR model is specifically an autoregressive model (AR model for short), which is a statistical method for processing time series. It uses the previous periods of the same variable, such as x, that is, X1 to x t-1 To predict this period X t , and assume that they are in a linear relationship. Because this is developed from linear regression in regression analysis, but instead of using X to predict y, it uses X to predict X (itself), so it is called autoregression.

[0138] The autocorrelation coefficient of the p-order autoregressive model is tailing, and the partial autocorrelation coefficient is truncated at p-order.

[0139] Autoregressive models are widely used in economics, informatics, and prediction of natural phenomena.

[0140]

[0141] Among them, c is a constant term, ε t is assumed to have a mean of 0, It is a fixed coefficient. The closer the time is to the future, the larger its value is. The standard deviation is equal to the random error value ε of σ t ; σ is assumed to be constant for any t, and the expected value of X is equal to the linear combination of one or more lag periods, plus a constant term, plus random errors.

[0142] 405, visualize the actual quantity and the prediction result of the prediction model, and calculate the error;

[0143] In this step-by-step, the prediction error value formula is: (AE) / E*100%, exceeding is positive, and under is negative.

[0144] A represents the measured value and E represents the normal value.

[0145] Forecast error rate calculation method:

[0146] a is the first measurement data, b is the second measurement data, c is the third measurement data, d is the fourth measurement data, and e is the fifth measurement data;

[0147] (a+b+c+d+e) / 5 = average value;

[0148] Average / 100 = percentage of average;

[0149] The error types include:

[0150] Model Error

[0151] In the process of establishing mathematical models, complex phenomena need to be abstracted and reduced to mathematical models, and the influence of some minor factors is often ignored and the problem is simplified. Therefore, there is a certain error between the mathematical model and the actual problem, which is called model error.

[0152] Measurement error

[0153] The data used in the modeling and specific calculation process are often obtained through observation and measurement. Due to the limitation of accuracy, these data are generally approximate, that is, there are errors, which are called measurement errors.

[0154] Truncation error

[0155] Since actual calculations can only complete a finite number of items or a finite number of steps, some calculations that require limits or infinite processes must be finite, and the infinite process must be truncated. The error generated in this way is called truncation error.

[0156] Rounding error

[0157] In the process of numerical calculation, due to the limitations of computing tools, some numbers are often rounded off, and only the first few digits are retained as the approximate value of the number. The error caused by rounding is called rounding error.

[0158] 406 , using the prediction model to predict the quantity of items in the determined time period to obtain the predicted quantity of items.

[0159] In this step, specifically, through the error calculation in step 405, prediction indicators, prediction models, and exponential smoothing times with lower errors are screened out for use in prediction, so as to improve the prediction accuracy.

[0160] The above describes the express delivery volume prediction method based on the AR model in the embodiment of the present invention. The following describes the express delivery volume prediction device based on the AR model in the embodiment of the present invention. Please refer to Figure 5 An embodiment of the express delivery quantity prediction device based on the AR model in the embodiment of the present invention includes:

[0161] The acquisition module 501 is used to read the historical data of the express delivery volume from the background database in the express delivery management platform, and extract the target data set of at least two historical periods from the historical data;

[0162] A difference module 502 is used to perform difference processing on at least two target data sets using a difference method to obtain a plurality of difference values, and optimize at least two target data sets based on the difference values ​​to obtain a sample data set;

[0163] The model generation module 503 is used to perform fitting training and learning on the plurality of obtained differential values ​​and the sample data set using a preset AR model algorithm to obtain a piece quantity prediction model;

[0164] A calculation module 504 is used to determine a cycle time to be predicted, and calculate corresponding cycle parameters and a historical target data set based on the cycle time;

[0165] The prediction module 505 is used to predict the express delivery quantity within the cycle time using the quantity prediction model according to the cycle parameters and the historical target data set to obtain the predicted quantity.

[0166] In the embodiment of the present invention, the express delivery quantity prediction device obtains the fluctuation coefficient by performing differential calculation on the acquired historical data, forms sample data based on the fluctuation coefficient, trains the model according to the AR model algorithm and the sample data, generates a quantity prediction model, calculates the corresponding cycle parameters and the historical target data set after determining the cycle time to be predicted, and uses the quantity prediction model to perform prediction to obtain the predicted quantity of the cycle time. Using differentials to generate samples improves the stability of the samples, making the obtained model more accurate and improving the accuracy of the predicted quantity. At the same time, through AR model prediction, manual operations are reduced, thus improving work efficiency.

[0167] See also Figure 6 Another embodiment of the express delivery quantity prediction device based on the AR model in the embodiment of the present invention includes:

[0168] The acquisition module 501 is used to read the historical data of the express delivery volume from the background database in the express delivery management platform, and extract the target data set of at least two historical periods from the historical data;

[0169] A difference module 502 is used to perform difference processing on at least two target data sets using a difference method to obtain a plurality of difference values, and optimize at least two target data sets based on the difference values ​​to obtain a sample data set;

[0170] The model generation module 503 is used to perform fitting training and learning on the plurality of obtained differential values ​​and the sample data set using a preset AR model algorithm to obtain a piece quantity prediction model;

[0171] A calculation module 504 is used to determine a cycle time to be predicted, and calculate corresponding cycle parameters and a historical target data set based on the cycle time;

[0172] The prediction module 505 is used to predict the express delivery quantity within the cycle time using the quantity prediction model according to the cycle parameters and the historical target data set to obtain the predicted quantity.

[0173] In this embodiment, the acquisition module 501 is specifically used for:

[0174] Extracting waybills from the historical data, and scanning and identifying the barcodes in each waybill to determine the type of each waybill and the processing time of the waybill, wherein the types include receiving waybills and sending waybills;

[0175] Classifying the historical data according to the types to obtain a historical receipt data set and a historical outgoing data set;

[0176] According to the preset data filtering rules, abnormal data and redundant information are respectively identified for the historical received data set and the historical sent data set, and corresponding identification information is marked to obtain a corresponding identification data set;

[0177] Correcting corresponding data or information from a corresponding identification data set according to the identification information, and filling blank data in the corrected data set by replacing adjacent data to obtain a new data set;

[0178] Identify the order number information and address information in the new data set, and delete them to obtain a first historical receipt data set and a first historical delivery data set;

[0179] The first historical received data set and the first historical sent data set are respectively merged and divided into daily periods according to the processing time to obtain target data sets of at least two historical periods, wherein the target data sets include at least one waybill data.

[0180] In this embodiment, the acquisition module 501 is specifically used for:

[0181] According to the processing time, the waybill data in the first historical receiving data set and the first historical sending data set are sorted in chronological order to obtain a historical receiving data sequence and a historical sending data sequence;

[0182] Merging the waybill data in the historical receiving data sequence and the historical sending data sequence with the same processing time on a daily basis to form at least two small data sets, wherein the small data set includes at least one of the receiving waybill and the sending waybill;

[0183] The total quantity of the receiving waybills and / or the sending waybills in each small data set is counted and marked to generate at least two target data sets.

[0184] In this embodiment, the difference module 502 includes:

[0185] A logarithmic unit 5021 is used to perform multiple logarithmic calculations on the total number of pieces of each target data set to obtain multiple logarithmic values ​​of each target data set;

[0186] The difference unit 5022 is used to perform difference calculation on the multiple logarithmic values ​​by using a difference method to obtain multiple difference values, wherein the difference calculation is to subtract the same logarithmic values ​​of the total number of pieces of two consecutive target data sets;

[0187] A judging unit 5023 is used to judge whether a plurality of differential values ​​corresponding to each target data set meets a preset threshold;

[0188] The sample generating unit 5024 is used to remove the target data set corresponding to the fluctuation coefficient that does not meet the threshold value when it is determined that the difference value does not meet the threshold value, so as to obtain a sample data set.

[0189] In this embodiment, the difference value includes a first-order difference value and a second-order difference value, and the model generation module 503 includes:

[0190] The first calculation unit 5031 is used to perform self-regression calculation on the waybill data in the sample data set to obtain corresponding regression variables;

[0191] A regression unit 5032 is used to perform a random linear combination on the waybill data according to each regression variable, and perform regression analysis on the linear combination to obtain an autocorrelation coefficient between each waybill data;

[0192] The training unit 5033 is used to output the sample data set to a function formula corresponding to a preset AR model algorithm, and perform fitting training of the sample based on the autocorrelation coefficient and the second-order difference value in the difference value to obtain model parameters;

[0193] The generating unit 5034 is used to adjust the functional formula based on the model parameters to obtain a piece quantity prediction model.

[0194] In this embodiment, the express delivery quantity prediction device based on the AR model further includes an error calculation module 506, which is specifically used to:

[0195] Select at least two waybill data from the sample data set to form a test data set; input the test data set into the piece quantity prediction model to predict the piece quantity and obtain a predicted value;

[0196] Calculate the error between the predicted value and the waybill data in the test data set;

[0197] Determining whether the error is within a preset error range;

[0198] If not, filter the waybill data whose error is not within the error position to form a new sample data set;

[0199] The piece quantity prediction model is trained a second time based on the new sample data to output an optimal piece quantity prediction model.

[0200] In this embodiment, the calculation module 504 includes:

[0201] A screening unit 5041 is used to screen out waybill data whose processing time is close to the cycle time from at least two of the target data sets to form a historical target data set;

[0202] A calculation unit 5042, used to calculate the correlation coefficient between each waybill data and the processing time in the historical target data set;

[0203] The determination unit 5043 is used to determine the autocorrelation coefficient of each waybill data in the historical target data set according to the correlation coefficient to obtain a period parameter, wherein the period parameter is a coefficient sequence after the autocorrelation coefficients are sorted.

[0204] Through the implementation of the above method, the differential calculation is performed on the acquired historical data to obtain the differential value, and the historical data is formed into sample data based on the differential value. The model is trained according to the AR model algorithm, the differential value and the sample data to generate a piece quantity prediction model. After determining the cycle time to be predicted, the corresponding cycle parameters and the historical target data set are calculated, and the piece quantity prediction model is used to perform prediction to obtain the predicted piece quantity of the cycle time. This prediction method not only improves the accuracy of the prediction, but also can predict the piece quantity in advance, prepare people and vehicles for physical examinations, reduce costs, and reduce losses.

[0205] See also Figure 7 , the following describes in detail an embodiment of an express delivery quantity prediction device based on an AR model in an embodiment of the present invention from the perspective of hardware processing.

[0206] Figure 7 It is a structural schematic diagram of a courier quantity prediction device based on an AR model provided by an embodiment of the present invention. The courier quantity prediction device 700 based on the AR model may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 (for example, one or more mass storage devices) storing application programs 733 or data 732. Among them, the memory 720 and the storage medium 730 can be short-term storage or permanent storage. The program stored in the storage medium 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the route-based user demand prediction device 700. Furthermore, the processor 710 can be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the courier quantity prediction device 700 based on the AR model.

[0207] The express delivery quantity prediction device 700 based on the AR model may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input and output interfaces 760, and / or one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It can be understood by those skilled in the art that Figure 7 The structure of the express delivery volume prediction device based on the AR model shown does not constitute a limitation of the express delivery volume prediction device based on the AR model, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0208] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the express delivery volume prediction method based on an AR model.

[0209] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0210] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0211] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the quantity of express parcels based on an AR model, applied to an express delivery and receiving management platform, characterized in that: The express delivery volume prediction method comprises: Read historical data of express delivery volume from the background database in the express delivery management platform, and extract target data sets of at least two historical periods from the historical data; Performing differential processing on at least two of the target data sets using a differential method to obtain a plurality of differential values, and optimizing at least two of the target data sets based on the differential values ​​to obtain a sample data set; Using a preset AR model algorithm to perform fitting training and learning on the plurality of differential values ​​and the sample data set to obtain a piece quantity prediction model; Determine a cycle time to be predicted, and calculate corresponding cycle parameters and a historical target data set based on the cycle time; According to the cycle parameters and the historical target data set, the express delivery quantity within the cycle time is predicted using the quantity prediction model to obtain a predicted quantity; The extracting of at least two historical period target data sets from the historical data comprises: Extracting waybills from the historical data, and scanning and identifying the barcodes in each waybill to determine the type of each waybill and the processing time of the waybill, wherein the types include receiving waybills and sending waybills; Classifying the historical data according to the types to obtain a historical receipt data set and a historical outgoing data set; According to the preset data filtering rules, abnormal data and redundant information are respectively identified for the historical received data set and the historical sent data set, and corresponding identification information is marked to obtain a corresponding identification data set; Correcting corresponding data or information from a corresponding identification data set according to the identification information, and filling blank data in the corrected data set by replacing adjacent data to obtain a new data set; Identify the order number information and address information in the new data set, and delete them to obtain a first historical receipt data set and a first historical delivery data set; The first historical received data set and the first historical sent data set are respectively merged and divided into daily periods according to the processing time to obtain target data sets of at least two historical periods, wherein the target data sets include at least one waybill data.

2. The method for predicting the express delivery volume based on the AR model according to claim 1, characterized in that: The historical received data set and the historical sent data set are merged and divided into daily periods according to the processing time to obtain target data sets of at least two historical periods, including: According to the processing time, the waybill data in the first historical receiving data set and the first historical sending data set are sorted in chronological order to obtain a historical receiving data sequence and a historical sending data sequence; Merging the waybill data in the historical receiving data sequence and the historical sending data sequence with the same processing time on a daily basis to form at least two small data sets, wherein the small data set includes at least one of the receiving waybill and the sending waybill; The total quantity of the receiving waybills and / or the sending waybills in each small data set is counted and marked to generate at least two target data sets.

3. The method for predicting the express delivery volume based on the AR model according to claim 2, characterized in that: The step of performing differential processing on at least two target data sets by using a differential method to obtain a plurality of differential values, and optimizing at least two target data sets based on the differential values ​​to obtain a sample data set includes: Perform multiple logarithmic calculations on the total number of pieces of each target data set to obtain multiple logarithmic values ​​of each target data set; Performing differential calculation on the multiple logarithmic values ​​by using a differential method to obtain multiple differential values, wherein the differential calculation is to subtract the same logarithmic values ​​of the total number of pieces of two consecutive target data sets; Determine whether multiple differential values ​​corresponding to each target data set meet a preset threshold; If not, the target data set corresponding to the difference value that does not meet the threshold is eliminated to obtain a sample data set.

4. The method for predicting the express delivery volume based on the AR model according to any one of claims 1 to 3, characterized in that: The differential value includes a first-order differential value and a second-order differential value. The method of using a preset AR model algorithm to perform fitting training and learning on a plurality of the differential values ​​and the sample data set to obtain a piece quantity prediction model includes: Performing self-regression calculation on the waybill data in the sample data set to obtain corresponding regression variables; According to each of the regression variables, the waybill data are randomly linearly combined, and the linear combination is subjected to regression analysis to obtain the autocorrelation coefficient between each of the waybill data; Outputting the sample data set to a function formula corresponding to a preset AR model algorithm, and performing sample fitting training based on the autocorrelation coefficient and the second-order difference value in the difference value to obtain model parameters; The functional formula is adjusted based on the model parameters to obtain a piece quantity prediction model.

5. The method for predicting the express delivery volume based on the AR model according to any one of claims 1 to 3, characterized in that: After the preset AR model algorithm is used to perform fitting training and learning on the plurality of differential values ​​and the sample data set to obtain the piece quantity prediction model, the method further includes: Select at least two waybill data from the sample data set to form a test data set; Inputting the test data set into the piece quantity prediction model to predict the piece quantity and obtain a predicted value; Calculate the error between the predicted value and the waybill data in the test data set; Determining whether the error is within a preset error range; If not, filter the waybill data whose error is not within the error position to form a new sample data set; The piece quantity prediction model is trained a second time based on the new sample data to output an optimal piece quantity prediction model.

6. The method for predicting the express delivery volume based on the AR model according to claim 5, characterized in that: The calculating of the corresponding cycle parameters and historical target data set based on the cycle time includes: Filtering out waybill data whose processing time is close to the cycle time from at least two of the target data sets to form a historical target data set; Calculate the correlation coefficient between each waybill data and the processing time in the historical target data set; The autocorrelation coefficient of each waybill data in the historical target data set is determined according to the correlation coefficient to obtain a period parameter, wherein the period parameter is a coefficient sequence after the autocorrelation coefficients are sorted.

7. A device for predicting the quantity of express parcels based on an AR model, characterized in that: The express delivery quantity prediction device comprises: A collection module, used to read the historical data of the express delivery volume from the background database in the express delivery and receiving management platform, and extract the target data set of at least two historical periods from the historical data; A difference module, used for performing difference processing on at least two of the target data sets using a difference method to obtain a plurality of difference values, and optimizing at least two of the target data sets based on the difference values ​​to obtain a sample data set; A model generation module, used to perform fitting training and learning on a plurality of obtained differential values ​​and the sample data set using a preset AR model algorithm to obtain a piece quantity prediction model; A calculation module, used to determine the cycle time to be predicted, and calculate the corresponding cycle parameters and historical target data set based on the cycle time; A prediction module, configured to predict the express delivery quantity within the cycle time by using the quantity prediction model according to the cycle parameters and the historical target data set, so as to obtain a predicted quantity; The acquisition module is specifically used to: extract the waybills in the historical data, scan and identify the barcodes in each waybill, determine the type of each waybill and the processing time of the waybill, wherein the types include receiving waybills and sending waybills; classify the historical data according to the types to obtain a historical receiving data set and a historical sending data set; identify abnormal data and redundant information in the historical receiving data set and the historical sending data set according to a preset data filtering rule, and mark them with corresponding identification information to obtain a corresponding identification data set; correct the corresponding data or information from the corresponding identification data set according to the identification information, and fill the blank data in the corrected data set by replacing adjacent data to obtain a new data set; identify the order number information and address information in the new data set, and delete them to obtain a first historical receiving data set and a first historical sending data set; merge and divide the first historical receiving data set and the first historical sending data set according to the processing time by daily periods to obtain target data sets of at least two historical periods, wherein the target data set includes at least one waybill data.

8. The express delivery volume prediction device based on AR model according to claim 7, characterized in that: The acquisition module is specifically used to: sort the waybill data in the first historical receiving data set and the first historical sending data set in chronological order according to the processing time to obtain a historical receiving data sequence and a historical sending data sequence; merge the waybill data with the same processing time in the historical receiving data sequence and the historical sending data sequence according to a daily cycle to form at least two small data sets, wherein the small data set includes at least one of a receiving waybill and a sending waybill; count the total number of receiving waybills and / or sending waybills in each small data set, mark the number of pieces, and generate at least two target data sets.

9. A device for predicting the quantity of express parcels based on an AR model, characterized in that: The express delivery quantity prediction device based on the AR model includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the express delivery volume prediction device based on the AR model to perform the steps of the express delivery volume prediction method based on the AR model as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the express delivery volume prediction method based on the AR model as described in any one of claims 1 to 6 are implemented.

Images