CN110019401B - Method, device, equipment and storage medium for predicting part quantity - Google Patents
Method, device, equipment and storage medium for predicting part quantity Download PDFInfo
- Publication number
- CN110019401B CN110019401B CN201711426164.0A CN201711426164A CN110019401B CN 110019401 B CN110019401 B CN 110019401B CN 201711426164 A CN201711426164 A CN 201711426164A CN 110019401 B CN110019401 B CN 110019401B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- training
- processing
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 139
- 238000012549 training Methods 0.000 claims abstract description 104
- 230000009467 reduction Effects 0.000 claims abstract description 15
- 238000012795 verification Methods 0.000 claims description 42
- 230000002776 aggregation Effects 0.000 claims description 33
- 238000004220 aggregation Methods 0.000 claims description 33
- 238000013136 deep learning model Methods 0.000 claims description 30
- 230000000737 periodic effect Effects 0.000 claims description 24
- 230000004931 aggregating effect Effects 0.000 claims description 12
- 238000010200 validation analysis Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 5
- 238000005192 partition Methods 0.000 claims 1
- 230000000379 polymerizing effect Effects 0.000 claims 1
- 238000000638 solvent extraction Methods 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000008859 change Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 13
- 238000013135 deep learning Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 9
- 230000015654 memory Effects 0.000 description 7
- 238000005096 rolling process Methods 0.000 description 7
- 238000011161 development Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000007787 long-term memory Effects 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 238000002360 preparation method Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012384 transportation and delivery Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2474—Sequence data queries, e.g. querying versioned data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2477—Temporal data queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Fuzzy Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a piece quantity prediction method, a piece quantity prediction device, piece quantity prediction equipment and a storage medium thereof. The method comprises the following steps: processing the historical waybill data and the external data within the first time range to obtain a data processing result; training by using the data processing result to obtain a depth residual error network basic model; and updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number. The technical scheme provided by the embodiment of the application overcomes the problem of overhigh maintenance cost caused by single net point independent modeling in the prior art; and through dimension reduction processing, period and trend learning processing on the historical waybill data, a depth residual error network model is obtained through training to predict the number of pieces of future waves, so that the accuracy of a prediction result is improved. The basic model is updated by scrolling the update data, thereby greatly reducing the consumption of computing resources and the time overhead.
Description
Technical Field
The present disclosure relates generally to the field of computers, and in particular, to the field of data mining processing, and more particularly, to a method, an apparatus, a device, and a storage medium for predicting a piece amount.
Background
The logistics field is an important tie for connecting social economic development and social life, the data mining based on the field belongs to the emerging research field, the development of big data technology brings new opportunities to the logistics industry, and the big data technology is reasonably applied to play a positive role in the aspects of management and decision making, customer relation maintenance, resource allocation and the like of the logistics industry.
In the prior art, recurrent neural networks (recurrent neural network, RNN) in time series models are applied in the logistics industry for predicting delivery volume information of a website. For example, the time sequence prediction method based on the generalized additive model can predict the daily piece quantity of each website, but the prediction mode has the problems of large data quantity, small piece quantity average value and large prediction error in wave-order piece quantity prediction. In addition, the existing prediction mode has the problems that the network point data are more, the data are sparse, the maintenance cost is too high and the like due to separate modeling. In addition, the model built based on the dot data can have the problems of large data jitter, unstable model performance, slow modeling speed, high time consumption and the like by considering the influence of factors such as holidays, extreme weather, time dynamic change and the like.
Therefore, a new prediction model is needed to solve the above problems.
Disclosure of Invention
In view of the foregoing drawbacks or shortcomings in the prior art, it is desirable to provide a scheme for predicting a target number of wave-order components based on a depth residual network.
In a first aspect, an embodiment of the present application provides a method for predicting a part quantity, including:
processing the historical waybill data and the external data within the first time range to obtain a data processing result;
training by using the data processing result to obtain a depth residual error network basic model;
and updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number.
In a second aspect, embodiments of the present application provide a piece amount prediction device, including:
the data processing unit is used for processing the historical waybill data and the external data in the first time range to obtain a data processing result;
a basic model unit is established and used for training by utilizing the data processing result to obtain a depth residual error network basic model;
and the updating prediction unit is used for updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number.
In a third aspect, an embodiment of the present application provides an apparatus, including a processor, a storage device;
the aforementioned storage means for storing one or more programs;
the one or more programs described above are executed by the processor described above, so that the processor described above implements the methods described in the embodiments of the present application.
In a fourth aspect, embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, the computer program implementing the method described in the embodiments of the present application when the computer program is executed by a processor.
According to the quantity prediction scheme provided by the embodiment of the application, the important attribute data related to the quantity prediction of the wave number is extracted through data processing, the LSTM model is utilized to learn the period and trend characteristics of the wave number from the historical waybill data, and the external data influencing the quantity of the wave number is utilized to train the depth residual error network basic model based on aggregation of the information, so that training time can be shortened, and model creation efficiency is improved. Furthermore, the basis model is updated in a rolling prediction mode to predict the target wave number of pieces, so that the future wave number of pieces can be accurately predicted. According to the embodiment of the application, the stability of the prediction result is improved by training a plurality of different update models to perform cross verification.
According to the technical scheme of the embodiment of the application, the problem that maintenance cost is too high due to single net point independent modeling in the prior art is solved, and the depth residual error network model is obtained through training after the dimension reduction processing, the period and the trend learning processing of historical waybill data so as to predict the quantity of the piece of the missing wave, so that the accuracy of a prediction result is improved. The basic model is updated by scrolling the update data, so that the consumption of computing resources and the time cost are greatly reduced.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the detailed description of non-limiting embodiments, made with reference to the following drawings, in which:
fig. 1 shows a flow chart of a method for predicting a quantity of a part, which is provided in an embodiment of the present application;
FIG. 2 is a schematic flow chart of a method for predicting a quantity of a part according to another embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a data dimension reduction process;
fig. 4 shows a schematic diagram of periodic law and recent trend learning;
FIG. 5 shows a schematic diagram of a single time series learning;
FIG. 6 shows a schematic diagram of a depth residual network base model;
fig. 7 shows a schematic structural diagram of a component amount predicting device provided in an embodiment of the present application;
Fig. 8 is a schematic structural view of a component amount predicting device according to another embodiment of the present application;
fig. 9 shows a schematic diagram of a computer system suitable for use in implementing the terminal device of the embodiments of the present application.
Detailed Description
The present application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be noted that, for convenience of description, only the portions related to the invention are shown in the drawings.
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
Referring to fig. 1, fig. 1 is a flow chart illustrating a method for predicting a quantity of a part according to an embodiment of the present application.
As shown in fig. 1, the method includes:
and step 101, processing the historical waybill data and the external data in the first time range to obtain a data processing result.
In the embodiment of the application, in a data preparation stage, historical waybill data and external data in a set time range are acquired, and the data are processed and then used for establishing a model. The data resources in the logistics industry are rich, and the historical waybill data can be extracted from the logistics data for processing. In order to solve the problems of large amount of dot data, small average value of the amount of pieces and large prediction error, a model can be created by all dot data of a city (or business area), and the amount of wave-order pieces of the dots can be predicted by unified modeling.
The set time range may be a first time range, and the value of the first time range may be set in units of years. For example, 2 years of historical waybill data. And acquiring historical waybill data of 2 years, extracting characteristic attribute data affecting the wave number of the wave number in the historical waybill data, and processing the characteristic attribute data to train a model. Wherein the feature attribute data includes: dot number data, month data, day data, week data, wave number data, and the like.
And performing dimension reduction processing on the characteristic attribute data, and extracting characteristic attributes with strong representativeness. The data after the dimension reduction processing is convenient for calculation and visualization, is more beneficial to extracting effective information and improves the efficiency of data processing.
Common ways of dimension reduction processing are linear mapping and non-linear mapping. For example, feature attribute data may be processed in an ebadd manner in a deep learning model to map to feature space.
According to the embodiment of the application, the historical waybill data is learned by using the deep learning model, so that a time sequence capable of reflecting the data development trend is obtained. The net point quantity data has stronger periodicity, and the regularity is extracted from the historical waybill data by using a deep learning model and is used for creating a basic model to predict the wave quantity of a future time period.
For example, the historical waybill data is learned by using a deep learning model, and a time sequence of a reaction cycle rule and a time sequence of a recent trend are extracted. A periodic time series, such as a time series in weeks, a trend time series, such as a 28 day trend time series. The deep learning model in the embodiment of the application can be a deep learning long-short-term memory model (LSTM) model, or other neural network models, and other models capable of being used for learning data rules. Preferably, embodiments of the present application select a deep learning long-term memory model (LSTM) model.
According to the embodiment of the application, the influence degree of external factors on the part quantity is considered, partial external data is also introduced in the modeling process, and the accuracy of the model is enhanced. Such as special date data, weather data, etc.
And step 102, training by using the data processing result to obtain a depth residual error network basic model.
In the embodiment of the application, the result after data processing is used for constructing the model. Before the training model of the data processed result is obtained, the data processed result is required to be subjected to aggregation processing, and the aggregation processed result is input into the constructed deep learning model for training. For example, feature attribute data subjected to the dimension reduction processing, a time sequence of a periodic rule and a time sequence of a recent trend, and external data are subjected to the aggregation processing, and a result of the aggregation processing is obtained. The results of the aggregation process are utilized to train a deep learning model. In the embodiment of the application, the depth residual network model is preferably taken as a basic model, and the basic model can comprise a plurality of residual modules. In the embodiment of the application, better results are obtained through a residual network by using fewer neural network layers, so that training time is shortened.
And step 103, updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number.
In the embodiment of the application, after the basic model is created, the basic model is retrained by using the updated data to obtain a new prediction model, so that the number of wave components in a future time period is predicted by using the prediction model. Optionally, the data is updated in a rolling prediction manner. For example, a part of the historical component amount data and a part of the newly acquired data are acquired to be combined into a training data set, and the basic model is updated month by month, so that the wave component amount data of a certain time period in the future is predicted.
The definition of the wave number can be understood as a work batch in a unit time, and the unit time can be one wave number every 30 minutes, and can be divided according to the work standard of the actual industry. The manner in which the wave times are processed is different according to the different categories of the waybill. For example, the types of the shipping notes may be divided into a dispatch class, an order receiving class, a no order receiving class, etc., wherein the wave number defined in the dispatch class divides 24 hours in the whole day into 35 wave numbers, the 1 st wave number is 0 to 7 in the early morning, the 2 nd wave number is 7 to 7:30, the 3 rd wave number is 7:30 to 8, and the like, a wave number is generated every half hour, and the like until the 35 th wave number is 23:30 to 0, and the wave number data of the dispatch class is processed to make the piece number more in line with the business scene, for example, the piece number data of each wave number from 21 to 9 in the morning in the previous day is accumulated to the 5 th wave number in the next day.
For another example, the wave number defined in the order receiving class is divided into 35 wave numbers in 24 hours on the whole day similar to the dispatch class, but the wave number of the order receiving class is processed, for example, the piece number data from the 31 st wave number on the previous day to the 4 th wave number on the next day is accumulated to the 4 th wave number on the next day. For another example, the no order receiving class, similar to the dispatch class, is divided into 35 wave times in 24 hours of the whole day, but the wave times of the no order receiving class are processed, for example, the piece quantity data from 35 wave times of the previous day to 3 wave times of the next day are uniformly distributed to 26 wave times to 34 wave times of the previous day.
According to the embodiment of the application, the prediction of the wave number is performed through deep learning, so that the good performance is realized, the number of the wave number of each wave number can be predicted accurately, and the prediction effect is greatly improved. According to the embodiment of the application, the deep learning long-short-term memory network (LSTM) model is adopted to mine the change modes in the time sequence, long-term change rules are considered, meanwhile, recent change trends are considered, and the model training time is shortened and the efficiency of model creation is improved by creating the depth residual error network basic model. The basic model is updated in a rolling prediction mode, so that the problems of low creation speed and high time consumption of the traditional time sequence model in the prior art are solved.
Referring to fig. 2, fig. 2 is a flow chart illustrating a method for predicting a quantity of a workpiece according to another embodiment of the disclosure.
As shown in fig. 2, the method includes:
and step 201, processing the historical waybill data and the external data in the first time range to obtain a data processing result.
In the embodiment of the application, in a data preparation stage, historical waybill data and external data in a set time range are acquired, and the data are processed and then used for establishing a model. The set time range may be a first time range, and the value of the first time range may be set in units of years.
For the historical waybill data, to obtain the characteristic attribute data affecting the wave number of the pieces of the waybill data, the characteristic attribute data needs to be extracted from the historical waybill data, and the characteristic attribute data is processed to obtain a first processing result.
For example, historical waybill data from 2015, month 2, to 2017, month 2 is taken as an example. And extracting characteristic attribute data affecting the wave number of the components from the historical waybill data of the period of time. Wherein the feature attribute data includes: dot number data, month data, day data, week data, wave number data, and the like. Alternatively, as shown in fig. 3, the dot number data, month data, day data, week data, and wave number data are respectively subjected to the ebadd processing, and then aggregated to generate the identification information, so as to realize the dimension reduction processing of the feature attribute data.
For historical waybill data, to acquire the data change rule characteristics affecting the wave number of the waybill, the historical waybill data needs to be subjected to periodic rule learning based on a deep learning model to obtain a plurality of first time sequences, and the plurality of first time sequences are aggregated into a periodic time sequence; and similarly, carrying out recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence. The deep learning model in the embodiment of the application can be a deep learning long-short-term memory model (LSTM) model, or other neural network models, and other models capable of being used for learning data rules. Preferably, embodiments of the present application select a deep learning long-term memory model (LSTM) model.
For example, for historical waybill data of 2015, 2 months-2017, 2 months, a time sequence is mined by using an LSTM model to reflect the periodic variation rule and the recent development trend rule of the data. As shown in fig. 4, the historical waybill data is input into the LSTM model, and the historical periodic time series is extracted and aggregated as the periodic time series. And inputting the historical waybill data into the LSTM model, and extracting a historical trend time sequence as a trend time sequence.
If the time sequence of a monday in the historical waybill data is extracted, the monday data of 12 weeks before the monday is extracted in a time backtracking mode based on the monday, and is respectively input into an LSTM module for processing, and then is output to a 32-layer full-connection layer for processing after being processed by 64-layer full-connection layers, so that the time sequence of the monday is generated, as shown in fig. 5.
Wherein the trend time series is extracted using historical waybill data, as shown in fig. 5, the first 28 days of data for a specified time may be extracted to generate the trend time series.
In the embodiment of the application, external data is introduced to further analyze the change of the wave number components in the data processing stage. And carrying out identification processing on the external data to obtain a second processing result.
There are various ways of acquiring external data. For example, extracted from internal data, or acquired through external data resources, etc. The acquired external data needs to be further processed and converted into characteristics affecting the wave number of pieces. The external data extracted in the embodiment of the present application may be special date data and weather data. Wherein the special dates include a portion of legal holidays and a portion of special dates, such as, for example, primordial denier, spring festival, noon, qing Ming, wuyi, mid-autumn, national celebration, double eleven, double twelve, etc. The processing of special date data can be understood as taking the element denier, the spring festival, the noon, the Qing Ming, the five elements, the mid-autumn, the national celebration, the double eleven and the double twelve as columns, the date data as rows, and the value range of the data in each column is [ -15, 16]. Taking the spring festival (namely the first year of the lunar calendar) as an example, 16 indicates that 15 days before and after the date are not spring festival, -15 indicates that 15 days are spring festival, 0 indicates that the spring festival is the same day, and 15 indicates that the spring festival has passed for 15 days. By carrying out identification processing on the special date, the influence relation between the special date and the wave number can be better mined.
For the processing of weather data it is understood that the weather is represented numerically, for example in a range of values of-2,0,2, where-2 represents extremely bad weather and 2 represents clear, uncertain or normal time candidate 0. And processing weather data to better analyze the influence of weather factors on the wave number of the wave number components.
And after the special date data and the weather data are processed, the external data are generated through aggregation.
Optionally, step 201 may further include:
step 2011, extracting characteristic attribute data from the historical waybill data, and processing the characteristic attribute data to obtain a first processing result;
step 2012, performing periodic rule learning on the historical waybill data based on the deep learning model to obtain a plurality of first time sequences, and aggregating the plurality of first time sequences into a periodic time sequence;
step 2013, performing recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence;
and step 2014, performing identification processing on the external data to obtain a second processing result, wherein the external data comprises special date data and/or weather data.
And 202, training by using the data processing result to obtain a depth residual error network basic model.
And processing the historical quantity data to obtain a data processing result, and aggregating the data processing result for training a depth residual error network basic model.
In the embodiment of the application, the first processing result, the period time sequence, the trend time sequence and the second processing result are aggregated to obtain an aggregation result. And inputting the aggregation result into a depth residual error network model for training. The depth residual network model may include a plurality of residual modules. As shown in fig. 6, the first processing result, the cycle time series, the trend time series, and the second processing result are input into the depth residual network base model in aggregate. The depth residual network base model includes 3 residual modules. Each residual error module is constructed by carrying out residual error learning on a full connection layer.
Optionally, step 202 includes:
step 2021, aggregating the data processing result to obtain an aggregation result;
and step 2022, training by using the aggregation result to obtain a depth residual error network basic model.
Wherein step 2021 further comprises: and aggregating the processing result, the plurality of time sequences, the second time sequence and the second processing result to obtain an aggregation result.
And 203, updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number.
The existing time series model trains each net point for a longer time per wave, generally greater than 20 minutes, with a wave number of about 30 waves per day. It can be seen that the existing time series model consumes too long training time, resulting in lower prediction efficiency of the model. In order to more effectively predict the change of the future wave number of components, a model with better updating performance needs to be constructed so as to improve the working efficiency.
In the embodiment of the application, a depth residual network base model is constructed in step 202, and the base model can be updated by setting a rolling prediction mode for the base model.
For example, the base model is updated with historical waybill data over a second time horizon. The setting of the second time range may be dynamically adjusted according to the data update status. For example, the predicted target is month 3 in 2017, then the second time range may be set to month 10 in 2016-month 2 in 2017. If the predicted objective is month 4 of 2017, the second time range may be set to month 11 of 2016-month 3 of 2017. Alternatively, further narrowing down the time range, for example, predicting that the target is 3 months in 2017, the second time range may be set to 12 months in 2016-2 months in 2017.
In this embodiment of the present application, the second time range is selected by scrolling in a month, and a range of the set time length is selected. The length of time may be an integer multiple of the natural month. For example, 3 months or 5 months.
After the historical waybill data for the second time range is acquired, the historical waybill data for the period of time is further divided into a training data set and a verification data set. And retraining the basic model by using the training data set to update the basic model, inputting the updated basic model based on the verification data set to obtain the predicted wave number, and preventing the updated model from having the fitting problem by dividing the training data set and the verification data set in a cross verification mode.
Step 203, optionally, includes:
step 2031, determining historical waybill data of a second time range of the predicted target wave times;
step 2032, dividing the historical waybill data of the second time range into a training data set and a verification data set;
step 2033, retraining the depth residual error network basic model by using the training data set to obtain a prediction model;
step 2034, inputting the verification data set into the prediction model to obtain a predicted piece quantity of the target wave number.
For example, taking the example of predicting the number of wave times of 3 months in 2017, the second time range is 5 months, namely, historical waybill data of 2016, 10 months, and 2017, 2 months is selected. The historical waybill data is divided into a training data set and a verification data set.
For example, {2016, 10-2017, 1 } is a training dataset and {2017, 2 } is a validation dataset;
training the basic model by using a training data set to obtain an updated basic model, wherein the updated basic model is called a prediction model for distinguishing; and inputting the verification data set into a prediction model to obtain a prediction result, namely the predicted piece quantity of the target wave number.
Optionally, in the embodiment of the present application, in order to further improve stability of a prediction result, cross-validation is performed by adjusting a training set and training multiple base models of a validation data set to implement prediction of a wave number component.
Optionally, step 2032 may further include:
in the historical waybill data of the second time range, a plurality of verification data sets are dynamically designated, and a plurality of training data sets corresponding to the verification data sets one by one are divided.
Optionally, step 2033 may further include:
and respectively inputting the plurality of training data sets into the depth residual error network basic model for retraining to obtain a plurality of depth residual error network prediction models corresponding to the training data sets one to one.
Optionally, step 2034 may further include:
respectively inputting verification data sets corresponding to the training data sets one by one into a depth residual error network model corresponding to the training data sets one by one to obtain a plurality of predicted piece quantities;
An average value of the plurality of predicted pieces of quantity is calculated as the predicted piece quantity of the target wave number.
For example, taking the example of predicting the number of wave times of 3 months in 2017, the second time range is 5 months, namely, historical waybill data of 2016, 10 months, and 2017, 2 months is selected.
Historical waybill data from month 2016, 10, to month 2017, 2, was divided into 3 different training data sets and verification data sets.
For example, {2016, 10-2017, 1 } is the first training dataset and {2017, 2 } is the first validation dataset;
{2016 month 12-2016 month 10, 2017 month 2 } is the second training dataset, and {2017 month 1 } is the second validation dataset;
{2016 month 11-10 months, 2017 month 1-2 months } is the third training dataset, and {2016 month 12 } is the third validation dataset.
Training the basic model by using a first training data set to obtain an updated basic model, wherein the updated basic model is called a first prediction model for distinguishing expression; and inputting the first verification data set into a first prediction model to obtain a first prediction result.
Training the basic model by using a second training data set to obtain an updated basic model, wherein the updated basic model is called a second prediction model for distinguishing expression; and inputting the first verification data set into a second prediction model to obtain a second prediction result.
Training the basic model by using a third training data set to obtain an updated basic model, wherein the updated basic model is called a third prediction model for distinguishing expression; and inputting the first verification data set into the second prediction model to obtain a third prediction result.
And finally, averaging the first prediction result, the second prediction result and the second prediction result to obtain a final prediction result.
It should be noted that although the operations of the method of the present invention are depicted in the drawings in a particular order, this does not require or imply that the operations must be performed in that particular order or that all of the illustrated operations be performed in order to achieve desirable results. Rather, the steps depicted in the flowcharts may change the order of execution. For example, a step of performing periodic rule learning on historical waybill data based on a deep learning model to obtain a plurality of first time sequences, and aggregating the plurality of first time sequences into a periodic time sequence, and a step of performing recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform. For example, training is performed by using the data processing result to obtain a depth residual network basic model, which includes: aggregating the data processing result to obtain an aggregation result; training by using the aggregation result to obtain a depth residual error network basic model.
Referring to fig. 7, fig. 7 is a schematic structural diagram of a part quantity predicting device according to an embodiment of the present application.
As shown in fig. 7, the apparatus 700 includes:
the data processing unit 701 is configured to process the historical waybill data and the external data within the first time range to obtain a data processing result.
In the embodiment of the application, in a data preparation stage, historical waybill data and external data in a set time range are acquired, and the data are processed and then used for establishing a model. The data resources in the logistics industry are rich, and the historical waybill data can be extracted from the logistics data for processing. To overcome the problem of sparse dot data, a model can be created by using all dot data of a city (or business area), and the wave number of dots can be predicted by unified modeling.
The set time range may be a first time range, and the value of the first time range may be set in units of years. For example, 2 years of historical waybill data. And acquiring historical waybill data of 2 years, extracting characteristic attribute data affecting the wave number of the wave number in the historical waybill data, and processing the characteristic attribute data to train a model. Wherein the feature attribute data includes: dot number data, month data, day data, week data, wave number data, and the like.
And performing dimension reduction processing on the characteristic attribute data, and extracting characteristic attributes with strong representativeness. The data after the dimension reduction processing is convenient for calculation and visualization, is more beneficial to extracting effective information and improves the efficiency of data processing.
Common ways of dimension reduction processing are linear mapping and non-linear mapping. For example, feature attribute data may be processed in an ebadd manner in a deep learning model to map to feature space.
According to the embodiment of the application, the historical waybill data is learned by using the deep learning model, so that a time sequence capable of reflecting the data development trend is obtained. The net point quantity data has stronger periodicity, and the regularity is extracted from the historical waybill data by using a deep learning model and is used for creating a basic model to predict the wave quantity of a future time period.
For example, the historical waybill data is learned by using a deep learning model, and a time sequence of a reaction cycle rule and a time sequence of a recent trend are extracted. A periodic regular time sequence, such as a time sequence in weeks, a time sequence of recent trends, such as a time sequence in months. The deep learning model in the embodiment of the application can be a deep learning long-short-term memory model (LSTM) model, or other neural network models, and other models capable of being used for learning data rules. Preferably, embodiments of the present application select a deep learning long-term memory model (LSTM) model.
According to the embodiment of the application, the influence degree of external factors on the part quantity is considered, partial external data is also introduced in the modeling process, and the accuracy of the model is enhanced. Such as special date data, weather data, etc.
A basic model unit 702 is created, and is used for training by using the data processing result to obtain a depth residual error network basic model.
In the embodiment of the application, the result after data processing is used for constructing the model. Before the training model of the data processed result is obtained, the data processed result is required to be subjected to aggregation processing, and the aggregation processed result is input into the constructed deep learning model for training. For example, feature attribute data subjected to the dimension reduction processing, a time sequence of a periodic rule and a time sequence of a recent trend, and external data are subjected to the aggregation processing, and a result of the aggregation processing is obtained. The results of the aggregation process are utilized to train a deep learning model. In the embodiment of the application, the depth residual network model is preferably used as a base model, and the base model can comprise a plurality of residual modules, so that better results are obtained through a residual network by using fewer neural network layers, and training time is reduced.
The updating prediction unit 703 is configured to update the base model with the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece amount of the target wave number.
In the embodiment of the application, after the basic model is created, the basic model is retrained by using the updated data to obtain a new prediction model, so that the number of wave components in a future time period is predicted by using the prediction model. Optionally, the data is updated in a rolling prediction manner. For example, a part of the historical component amount data and a part of the newly acquired data are acquired to be combined into a training data set, and the basic model is updated month by month, so that the wave component amount data of a certain time period in the future is predicted.
The definition of the wave number can be understood as a work batch in a unit time, and the unit time can be one wave number every 30 minutes, and can be divided according to the work standard of the actual industry. The manner in which the wave times are processed is different according to the different categories of the waybill. For example, the types of the shipping notes may be divided into a dispatch class, an order receiving class, a no order receiving class, etc., wherein the wave number defined in the dispatch class divides 24 hours in the whole day into 35 wave numbers, the 1 st wave number is 0 to 7 in the early morning, the 2 nd wave number is 7 to 7:30, the 3 rd wave number is 7:30 to 8, and the like, a wave number is generated every half hour, and the like until the 35 th wave number is 23:30 to 0, and the wave number data of the dispatch class is processed to make the piece number more in line with the business scene, for example, the piece number data of each wave number from 21 to 9 in the morning in the previous day is accumulated to the 5 th wave number in the next day.
For another example, the wave number defined in the order receiving class is divided into 35 wave numbers in 24 hours on the whole day similar to the dispatch class, but the wave number of the order receiving class is processed, for example, the piece number data from the 31 st wave number on the previous day to the 4 th wave number on the next day is accumulated to the 4 th wave number on the next day. For another example, the no order receiving class, similar to the dispatch class, is divided into 35 wave times in 24 hours of the whole day, but the wave times of the no order receiving class are processed, for example, the piece quantity data from 35 wave times of the previous day to 3 wave times of the next day are uniformly distributed to 26 wave times to 34 wave times of the previous day.
According to the embodiment of the application, the prediction of the wave number is performed through deep learning, so that the good performance is realized, the number of the wave number of each wave number can be predicted accurately, and the prediction effect is greatly improved. According to the embodiment of the application, the deep learning long-short-term memory network (LSTM) model is adopted to mine the change modes in the time sequence, long-term change rules are considered, meanwhile, recent change trends are considered, and the model training time is shortened and the efficiency of model creation is improved by creating the depth residual error network basic model. The basic model is updated in a rolling prediction mode, so that the problems of low creation speed and high time consumption of the traditional time sequence model in the prior art are solved.
Referring to fig. 8, fig. 8 is a schematic structural diagram of a component amount predicting device according to another embodiment of the present disclosure.
As shown in fig. 8, the apparatus 800 includes:
the data processing unit 801 is configured to process the historical waybill data and the external data within the first time range to obtain a data processing result.
In the embodiment of the application, in a data preparation stage, historical waybill data and external data in a set time range are acquired, and the data are processed and then used for establishing a model. The set time range may be a first time range, and the value of the first time range may be set in units of years.
For the historical waybill data, to obtain the characteristic attribute data affecting the wave number of the pieces of the waybill data, the characteristic attribute data needs to be extracted from the historical waybill data, and the characteristic attribute data is processed to obtain a first processing result.
For example, historical waybill data from 2015, month 2, to 2017, month 2 is taken as an example. And extracting characteristic attribute data affecting the wave number of the components from the historical waybill data of the period of time. Wherein the feature attribute data includes: dot number data, month data, day data, week data, wave number data, and the like. Alternatively, as shown in fig. 3, the dot number data, month data, day data, week data, and wave number data are respectively subjected to the ebadd processing, and then aggregated to generate the identification information, so as to realize the dimension reduction processing of the feature attribute data.
For historical waybill data, to acquire the data change rule characteristics affecting the wave number of the waybill, the historical waybill data needs to be subjected to periodic rule learning based on a deep learning model to obtain a plurality of first time sequences, and the plurality of first time sequences are aggregated into a periodic time sequence; and similarly, carrying out recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence. The deep learning model in the embodiment of the application can be a deep learning long-short-term memory model (LSTM) model, or other neural network models, and other models capable of being used for learning data rules. Preferably, embodiments of the present application select a deep learning long-term memory model (LSTM) model.
For example, for historical waybill data of 2015, 2 months-2017, 2 months, a time sequence is mined by using an LSTM model to reflect the periodic variation rule and the recent development trend rule of the data. As shown in fig. 4, the history waybill data is input into the LSTM model, and the history period time series is extracted and aggregated as period information. And inputting the historical waybill data into the LSTM model, and extracting a historical trend time sequence as trend information.
If the time sequence of a monday in the historical waybill data is extracted, the monday data of 12 weeks before the monday is extracted in a time backtracking mode based on the monday, and is respectively input into an LSTM module for processing, and then is output to a 32-layer full-connection layer for processing after being processed by 64-layer full-connection layers, so that the time sequence of the monday is generated, as shown in fig. 5.
In the embodiment of the application, external data is introduced to further analyze the change of the wave number components in the data processing stage. And carrying out identification processing on the external data to obtain a second processing result.
There are various ways of obtaining external data. For example, extracted from internal data, or acquired through external data resources, etc. The acquired external data needs to be further processed and converted into characteristics affecting the wave number of pieces. The external data extracted in the embodiment of the present application may be special date data and weather data. Wherein the special dates include a portion of legal holidays and a portion of special dates, such as, for example, primordial denier, spring festival, noon, qing Ming, wuyi, mid-autumn, national celebration, double eleven, double twelve, etc. The processing of special date data can be understood as taking the element denier, the spring festival, the noon, the Qing Ming, the five elements, the mid-autumn, the national celebration, the double eleven and the double twelve as columns, the date data as rows, and the value range of the data in each column is [ -15, 16]. Taking the spring festival (namely the first year of the lunar calendar) as an example, 16 indicates that 15 days before and after the date are not spring festival, -15 indicates that 15 days are spring festival, 0 indicates that the spring festival is the same day, and 15 indicates that the spring festival has passed for 15 days. By carrying out identification processing on the special date, the influence relation between the special date and the wave number can be better mined.
For the processing of weather data it is understood that the weather is represented numerically, for example in a range of values of-2,0,2, where-2 represents extremely bad weather and 2 represents clear, uncertain or normal time candidate 0. And processing weather data to better analyze the influence of weather factors on the wave number of the wave number components.
And after the special date data and the weather data are processed, the external information is generated through aggregation.
Optionally, the data processing unit 801 may further include:
the first processing subunit 8011 is configured to extract feature attribute data from the historical waybill data, and process the feature attribute data to obtain a first processing result;
the first time sequence generating subunit 8012 is configured to perform periodic rule learning on the historical waybill data based on the deep learning model to obtain a plurality of first time sequences, and aggregate the plurality of first time sequences into a periodic time sequence;
the second time sequence generating subunit 8013 is configured to perform recent trend learning on the historical waybill data based on the deep learning model, so as to obtain a trend time sequence;
the second processing subunit 8014 performs identification processing on external data, to obtain a second processing result, where the external data includes special date data and/or weather data.
A basic model unit 802 is created, and training is performed by using the data processing result, so as to obtain a depth residual error network basic model.
And processing the historical quantity data to obtain a data processing result, and aggregating the data processing result for training a depth residual error network basic model.
In the embodiment of the application, the first processing result, the period time sequence, the trend time sequence and the second processing result are aggregated to obtain an aggregation result. And inputting the aggregation result into a depth residual error network model for training. The depth residual network model may include a plurality of residual modules. As shown in fig. 6, the first processing result, the cycle time series, the trend time series, and the second processing result are input into the depth residual network base model in aggregate. The depth residual network base model includes 3 residual modules. Each residual error module is constructed by carrying out residual error learning on a full connection layer.
Optionally, creating the base model element 802 includes:
an aggregation subunit 8021, configured to aggregate the data processing result to obtain an aggregation result;
the training subunit 8022 is configured to perform training by using the aggregation result, so as to obtain a depth residual error network basic model.
Wherein the aggregation subunit 8021 is further configured to aggregate the processing result, the plurality of first time sequences, the second time sequences, and the second processing result to obtain an aggregate result.
The updating prediction unit 803 is configured to update the base model with the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece amount of the target wave number.
The existing time series model trains each net point for a longer time per wave, generally greater than 20 minutes, with a wave number of about 30 waves per day. It can be seen that the existing time series model consumes too long training time, resulting in lower prediction efficiency of the model. In order to more effectively predict the change of the future wave number of components, a model with better updating performance needs to be constructed so as to improve the working efficiency.
In the embodiment of the application, a depth residual network base model is constructed in step 202, and the base model can be updated by setting a rolling prediction mode for the base model.
For example, the base model is updated with historical waybill data over a second time horizon. The setting of the second time range may be dynamically adjusted according to the data update status. For example, the predicted target is month 3 in 2017, then the second time range may be set to month 10 in 2016-month 2 in 2017. If the predicted objective is month 4 of 2017, the second time range may be set to month 11 of 2016-month 3 of 2017. Alternatively, further narrowing down the time range, for example, predicting that the target is 3 months in 2017, the second time range may be set to 12 months in 2016-2 months in 2017.
In this embodiment of the present application, the second time range is selected by scrolling in a month, and a range of the set time length is selected. The length of time may be an integer multiple of the natural month. For example, 3 months or 5 months.
After the historical waybill data for the second time range is acquired, the historical waybill data for the period of time is further divided into a training data set and a verification data set. And retraining the basic model by using the training data set to update the basic model, inputting the updated basic model based on the verification data set to obtain the predicted wave number, and preventing the updated model from having the fitting problem by dividing the training data set and the verification data set in a cross verification mode.
The update prediction unit 803 optionally includes:
a determining subunit 8031 for determining historical waybill data of a second time range of the predicted target wavelength;
a data dividing subunit 8032 for dividing the historical waybill data of the second time range into a training data set and a verification data set;
an updating subunit 8033, configured to retrain the depth residual error network basic model by using the training data set, so as to obtain a prediction model;
the prediction subunit 8034 is configured to input the verification data set into a prediction model, so as to obtain a predicted piece quantity of the target wave number.
For example, taking the example of predicting the number of wave times of 3 months in 2017, the second time range is 5 months, namely, historical waybill data of 2016, 10 months, and 2017, 2 months is selected. The historical waybill data is divided into a training data set and a verification data set.
For example, {2016, 10-2017, 1 } is a training dataset and {2017, 2 } is a validation dataset;
training the basic model by using a training data set to obtain an updated basic model, wherein the updated basic model is called a prediction model for distinguishing; and inputting the verification data set into a prediction model to obtain a prediction result, namely the predicted piece quantity of the target wave number.
Optionally, in the embodiment of the present application, in order to further improve stability of a prediction result, cross-validation is performed by adjusting a training set and training multiple base models of a validation data set to implement prediction of a wave number component.
Optionally, the data dividing subunit 8032 is further configured to dynamically designate a plurality of verification data sets in the historical waybill data in the second time range, and divide a plurality of training data sets corresponding to the verification data sets one to one.
Optionally, the updating subunit 8033 is further configured to input the plurality of training data sets into the depth residual network basic model for retraining, so as to obtain a plurality of depth residual network prediction models corresponding to the training data sets one to one.
Optionally, the predictor unit 8034 further comprises:
the prediction subunit is used for respectively inputting verification data sets which are in one-to-one correspondence with the training data sets into the depth residual error network model which is in one-to-one correspondence with the training data sets to obtain a plurality of predicted piece quantities;
and a calculating subunit for calculating an average value of the plurality of predicted piece amounts as the predicted piece amount of the target wave number.
For example, taking the example of predicting the number of wave times of 3 months in 2017, the second time range is 5 months, namely, historical waybill data of 2016, 10 months, and 2017, 2 months is selected.
Historical waybill data from month 2016, 10, to month 2017, 2, was divided into 3 different training data sets and verification data sets.
For example, {2016, 10-2017, 1 } is the first training dataset and {2017, 2 } is the first validation dataset;
{2016 month 12-2016 month 10, 2017 month 2 } is the second training dataset, and {2017 month 1 } is the second validation dataset;
{2016 month 11-10 months, 2017 month 1-2 months } is the third training dataset, and {2016 month 12 } is the third validation dataset.
Training the basic model by using a first training data set to obtain an updated basic model, wherein the updated basic model is called a first prediction model for distinguishing expression; and inputting the first verification data set into a first prediction model to obtain a first prediction result.
Training the basic model by using a second training data set to obtain an updated basic model, wherein the updated basic model is called a second prediction model for distinguishing expression; and inputting the first verification data set into a second prediction model to obtain a second prediction result.
Training the basic model by using a third training data set to obtain an updated basic model, wherein the updated basic model is called a third prediction model for distinguishing expression; and inputting the first verification data set into the second prediction model to obtain a third prediction result.
And finally, averaging the first prediction result, the second prediction result and the second prediction result to obtain a final prediction result.
It should be understood that the elements or modules recited in apparatus 700 or 800 correspond to the various steps in the methods described with reference to fig. 1 or 2. Thus, the operations and features described above for the method are equally applicable to the apparatus 700 or 800 and the units contained therein, and are not described in detail herein. The apparatus 700 or 800 may be implemented in advance in a browser or other security application of the electronic device, or may be loaded into a browser or security application of the electronic device by means of downloading or the like. The respective units in the apparatus 700 or 800 may cooperate with units in an electronic device to implement aspects of embodiments of the present application.
Referring now to FIG. 9, there is illustrated a schematic diagram of a computer system 900 suitable for use in implementing the terminal device or server of embodiments of the present application.
As shown in fig. 9, the computer system 900 includes a Central Processing Unit (CPU) 901, which can execute various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 902 or a program loaded from a storage section 908 into a Random Access Memory (RAM) 903. In the RAM 903, various programs and data necessary for the operation of the system 900 are also stored. The CPU 901, ROM 902, and RAM 903 are connected to each other through a bus 904. An input/output (I/O) interface 905 is also connected to the bus 904.
The following components are connected to the I/O interface 905: an input section 906 including a keyboard, a mouse, and the like; an output portion 907 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage portion 908 including a hard disk or the like; and a communication section 909 including a network interface card such as a LAN card, a modem, or the like. The communication section 909 performs communication processing via a network such as the internet. The drive 910 is also connected to the I/O interface 905 as needed. A removable medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is installed as needed on the drive 910 so that a computer program read out therefrom is installed into the storage section 908 as needed.
In particular, according to embodiments of the present disclosure, the process described above with reference to fig. 1 or 2 may be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for performing the method of fig. 1 or 2. In such an embodiment, the computer program may be downloaded and installed from the network via the communication portion 909 and/or installed from the removable medium 911.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The units or modules described in the embodiments of the present application may be implemented by software, or may be implemented by hardware. The described units or modules may also be provided in a processor, for example, as: a processor includes a data processing unit, a create base model unit, and an update prediction unit. The names of these units or modules do not constitute a limitation of the unit or module itself in some cases, and for example, a data processing unit may also be described as "a unit for processing data".
As another aspect, the present application also provides a computer-readable storage medium, which may be a computer-readable storage medium contained in the apparatus described in the above embodiments; or may be a computer-readable storage medium, alone, that is not assembled into a device. The computer-readable storage medium stores one or more programs for use by one or more processors in performing the part quantity prediction methods described herein.
The foregoing description is only of the preferred embodiments of the present application and is presented as a description of the principles of the technology being utilized. It will be appreciated by persons skilled in the art that the scope of the invention referred to in this application is not limited to the specific combinations of features described above, but it is intended to cover other embodiments in which any combination of features described above or equivalents thereof is possible without departing from the spirit of the invention. Such as the above-described features and technical features having similar functions (but not limited to) disclosed in the present application are replaced with each other.
Claims (16)
1. A method of predicting a quantity of a piece, the method comprising:
processing the historical waybill data and the external data within the first time range to obtain a data processing result;
aggregating the data processing result to obtain an aggregation result;
training by utilizing the aggregation result to obtain a depth residual error network basic model;
updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number;
the method for processing the historical waybill data and the external data in the first time range to obtain a data processing result comprises the following steps:
extracting characteristic attribute data from the historical waybill data, and performing dimension reduction processing on the characteristic attribute data to obtain a first processing result;
performing periodic rule learning on the historical waybill data based on a deep learning model to obtain a plurality of first time sequences, and aggregating the plurality of first time sequences into a periodic time sequence;
performing recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence;
and carrying out identification processing on the external data to obtain a second processing result, wherein the external data comprises special date data and/or weather data.
2. The method of claim 1, wherein aggregating the data processing results to obtain an aggregate result comprises:
and polymerizing the first processing result, the period time sequence, the trend time sequence and the second processing to obtain an aggregation result.
3. The method of any of claims 1-2, wherein updating the base model with historical waybill data over a second time horizon to obtain a predictive model to determine a predicted quantity of the target wave-times comprises:
determining historical waybill data of a second time range of the predicted target wave times;
dividing the historical waybill data of the second time range into a training data set and a verification data set;
retraining the depth residual error network basic model by utilizing the training data set to obtain a prediction model;
and inputting the verification data set into the prediction model to obtain the predicted piece quantity of the target wave number.
4. A method according to claim 3, wherein the dividing the historical waybill data for the second time horizon into a training dataset and a verification dataset comprises:
and dynamically designating a plurality of first verification data sets in the historical waybill data in the second time range, and dividing a plurality of first training data sets which are in one-to-one correspondence with the first verification data sets.
5. The method of claim 4, wherein retraining the depth residual network base model with the training dataset results in a predictive model, comprising:
and respectively inputting the plurality of first training data sets into the depth residual error network basic model for retraining to obtain a plurality of depth residual error network prediction models corresponding to the first training data sets one to one.
6. The method of claim 5, wherein said inputting the validation dataset into the predictive model yields a predicted piece count for a target wave-number, comprising:
respectively inputting the first verification data sets which are in one-to-one correspondence with the first training data sets into a first depth residual error network model which is in one-to-one correspondence with the first training data sets to obtain a plurality of first predicted piece quantities;
and calculating an average value of the plurality of first predicted piece amounts as the predicted piece amount of the target wave number.
7. The method of any of claims 1-2 or 4-6, wherein the extracting feature attribute data from the historical waybill data and processing the feature attribute data comprises:
extracting feature attribute data from the historical waybill data, wherein the feature attribute data at least comprises: dot number data, month data, date data, week data, wave number data;
And performing dimension reduction processing on the characteristic attribute data.
8. The method of claim 7, wherein said dimension-reducing the feature attribute data comprises:
and mapping the characteristic attribute data to a characteristic space by adopting an ebedding mode.
9. A piece quantity predicting device, the device comprising:
the data processing unit is used for processing the historical waybill data and the external data in the first time range to obtain a data processing result;
a basic model unit is established and is used for training by utilizing the data processing result to obtain a depth residual error network basic model;
the updating prediction unit is used for updating the basic model by using the historical waybill data in the second time range to obtain a prediction model to determine the predicted piece quantity of the target wave number;
the data processing unit specifically comprises:
the first processing subunit is used for extracting characteristic attribute data from the historical waybill data, and performing dimension reduction processing on the characteristic attribute data to obtain a first processing result;
the first time sequence generation subunit is used for carrying out periodic rule learning on the historical waybill data based on a deep learning model to obtain a plurality of first time sequences, and aggregating the plurality of first time sequences into a periodic time sequence;
The second time sequence generation subunit is used for carrying out recent trend learning on the historical waybill data based on the deep learning model to obtain a trend time sequence;
the second processing subunit is used for carrying out identification processing on the external data to obtain a second processing result, wherein the external data comprises special date data and/or weather data;
wherein the creating a basic model unit comprises:
the aggregation subunit is used for aggregating the data processing result to obtain an aggregation result;
and the training subunit is used for training by utilizing the aggregation result to obtain a depth residual error network basic model.
10. The apparatus of claim 9, wherein the aggregation subunit is further configured to aggregate the first processing result, the periodic time sequence, the trend time sequence, and the second processing result to obtain an aggregate result.
11. The apparatus according to any one of claims 9-10, wherein the update prediction unit comprises:
a determining subunit, configured to determine historical waybill data of a second time range of the predicted target wavelength;
the data dividing subunit is used for dividing the historical waybill data in the second time range into a training data set and a verification data set;
An updating subunit, configured to retrain the depth residual error network basic model by using the training data set, so as to obtain a prediction model;
and the prediction subunit is used for inputting the verification data set into the prediction model to obtain the predicted piece quantity of the target wave number.
12. The apparatus of claim 11, wherein the data partitioning subunit is further configured to dynamically designate a plurality of first verification data sets in the historical waybill data for the second time range and partition a plurality of first training data sets that are one-to-one with the first verification data sets.
13. The apparatus of claim 11, wherein the updating subunit is further configured to input the plurality of first training data sets into the depth residual network base model for retraining, respectively, to obtain a plurality of depth residual network prediction models in one-to-one correspondence with the first training data sets.
14. The apparatus of claim 12, wherein the predictor unit comprises:
the first prediction subunit is used for respectively inputting the first verification data sets which are in one-to-one correspondence with the plurality of first training data sets into a first depth residual error network model which is in one-to-one correspondence with the plurality of first training data sets to obtain a plurality of first predicted piece amounts;
And the calculating subunit is used for calculating the average value of the plurality of first predicted piece amounts as the predicted piece amount of the target wave number.
15. A terminal device comprises a processor and a storage device; the method is characterized in that:
the storage device is used for storing one or more programs;
the one or more programs, when executed by the processor, cause the processor to implement the method of any of claims 1-8.
16. A computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711426164.0A CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711426164.0A CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110019401A CN110019401A (en) | 2019-07-16 |
CN110019401B true CN110019401B (en) | 2024-04-05 |
Family
ID=67187147
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711426164.0A Active CN110019401B (en) | 2017-12-25 | 2017-12-25 | Method, device, equipment and storage medium for predicting part quantity |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110019401B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112801327A (en) * | 2019-11-14 | 2021-05-14 | 顺丰科技有限公司 | Method, device, equipment and storage medium for predicting logistics flow and modeling thereof |
CN112862137A (en) * | 2019-11-27 | 2021-05-28 | 顺丰科技有限公司 | Method and device for predicting quantity, computer equipment and computer readable storage medium |
CN112907267A (en) * | 2019-12-03 | 2021-06-04 | 顺丰科技有限公司 | Method and device for predicting cargo quantity, computer equipment and storage medium |
CN110889560A (en) * | 2019-12-06 | 2020-03-17 | 西北工业大学 | Express delivery sequence prediction method with deep interpretability |
CN112948763B (en) * | 2019-12-11 | 2024-04-09 | 顺丰科技有限公司 | Piece quantity prediction method and device, electronic equipment and storage medium |
CN112990520A (en) * | 2019-12-13 | 2021-06-18 | 顺丰科技有限公司 | Mesh point connection quantity prediction method and device, computer equipment and storage medium |
CN112990526A (en) * | 2019-12-16 | 2021-06-18 | 顺丰科技有限公司 | Method and device for predicting logistics arrival quantity and storage medium |
CN113077069B (en) * | 2020-01-03 | 2023-06-13 | 顺丰科技有限公司 | Modeling method, device, equipment and storage medium for predicting shift number |
CN111401940B (en) * | 2020-03-05 | 2023-07-04 | 杭州网易再顾科技有限公司 | Feature prediction method, device, electronic equipment and storage medium |
CN112183832A (en) * | 2020-09-17 | 2021-01-05 | 上海东普信息科技有限公司 | Express pickup quantity prediction method, device, equipment and storage medium |
CN112418534B (en) * | 2020-11-26 | 2022-10-14 | 上海东普信息科技有限公司 | Method and device for predicting quantity of collected parts, electronic equipment and computer readable storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105427194A (en) * | 2015-12-21 | 2016-03-23 | 西安美林数据技术股份有限公司 | Method and device for electricity sales amount prediction based on random forest regression |
CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
US20170249534A1 (en) * | 2016-02-29 | 2017-08-31 | Fujitsu Limited | Method and apparatus for generating time series data sets for predictive analysis |
CN107316083A (en) * | 2017-07-04 | 2017-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for updating deep learning model |
CN107330522A (en) * | 2017-07-04 | 2017-11-07 | 北京百度网讯科技有限公司 | Method, apparatus and system for updating deep learning model |
-
2017
- 2017-12-25 CN CN201711426164.0A patent/CN110019401B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105427194A (en) * | 2015-12-21 | 2016-03-23 | 西安美林数据技术股份有限公司 | Method and device for electricity sales amount prediction based on random forest regression |
US20170249534A1 (en) * | 2016-02-29 | 2017-08-31 | Fujitsu Limited | Method and apparatus for generating time series data sets for predictive analysis |
CN106886846A (en) * | 2017-04-26 | 2017-06-23 | 中南大学 | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term |
CN107316083A (en) * | 2017-07-04 | 2017-11-03 | 北京百度网讯科技有限公司 | Method and apparatus for updating deep learning model |
CN107330522A (en) * | 2017-07-04 | 2017-11-07 | 北京百度网讯科技有限公司 | Method, apparatus and system for updating deep learning model |
Also Published As
Publication number | Publication date |
---|---|
CN110019401A (en) | 2019-07-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110019401B (en) | Method, device, equipment and storage medium for predicting part quantity | |
US20230049931A1 (en) | Method of training machine learning models for making simulated estimations | |
Perla et al. | Equilibrium imitation and growth | |
CN110400021B (en) | Bank branch cash usage prediction method and device | |
CN109961165B (en) | Method, device, equipment and storage medium for predicting part quantity | |
CN111159157A (en) | Method and device for indexing processing of enterprise report data | |
CN114971748B (en) | Prediction data generation method, model training method, computer device, and storage medium | |
US20200050982A1 (en) | Method and System for Predictive Modeling for Dynamically Scheduling Resource Allocation | |
CN103729796A (en) | Method and system for sample survey | |
CN111738504A (en) | Enterprise financial index fund amount prediction method and device, equipment and storage medium | |
CN113222057A (en) | Data prediction model training method, data prediction device, data prediction equipment and data prediction medium | |
CN113949644A (en) | Method and system for predicting website access amount based on LSTM-prophet | |
CN111738507A (en) | Bank clearing position fund payment amount prediction method, device, equipment and medium | |
US10445399B2 (en) | Forecast-model-aware data storage for time series data | |
Hu et al. | An attention-mechanism-based traffic flow prediction scheme for smart city | |
CN115545248A (en) | Target object prediction method, device, equipment and medium | |
CN113627692A (en) | Complaint amount prediction method, complaint amount prediction device, complaint amount prediction apparatus, and storage medium | |
CN111738506A (en) | Cash center cash stock usage amount prediction method and device, electronic device, and medium | |
CN113723775B (en) | Enterprise and industry operation risk assessment method based on power big data | |
CN117764638B (en) | Electricity selling data prediction method, system, equipment and storage medium for power supply enterprises | |
Karri | LSTM and SHAP for transparent and effective IoB systems in IoT environments: household power consumption | |
CN117539948B (en) | Service data retrieval method and device based on deep neural network | |
US11620565B1 (en) | System and method for enhanced distribution of data to compute nodes | |
CN117036008A (en) | Automatic modeling method and system for multi-source data | |
Nikishina | DESIGN OF INFORMATION SYSTEMANALYSIS OF DEMOGRAPHIC PROCESSES |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |