CN111932292A - Cigarette product sales prediction method and device based on deep neural network - Google Patents
Cigarette product sales prediction method and device based on deep neural network Download PDFInfo
- Publication number
- CN111932292A CN111932292A CN202010661727.XA CN202010661727A CN111932292A CN 111932292 A CN111932292 A CN 111932292A CN 202010661727 A CN202010661727 A CN 202010661727A CN 111932292 A CN111932292 A CN 111932292A
- Authority
- CN
- China
- Prior art keywords
- sales
- characteristic parameter
- neural network
- time sequence
- characteristic
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Accounting & Taxation (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a device for predicting cigarette product sales based on a deep neural network, wherein the method comprises the following steps: constructing a neural network model for predicting the sales of cigarette products; acquiring periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values, sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values; according to a fixed preset time sequence period, taking all characteristic parameter values as training samples and correspondingly inputting the training samples into each input neuron of the neural network model for model training until all training samples are trained, and obtaining a cigarette product sales prediction model; and inputting the characteristics extracted from the actual sales data of the current preset time sequence period into the cigarette product sales prediction model to obtain a sales prediction value of the next preset time sequence period. By implementing the method, the accuracy of the cigarette product sales prediction is improved, and the prediction result is more in line with the actual situation.
Description
Technical Field
The invention relates to the technical field of data mining and big data analysis, in particular to a method and a device for predicting the sales of cigarette products based on a deep neural network.
Background
The release of cigarette products is a very important basic work of tobacco commercial companies, and the cigarette product sales orders brought by the release of the products directly influence the economic benefits of the commercial companies. The essence of cigarette release is to carry out sales volume prejudgment and strategic guidance on cigarette sales work at the next stage, the release strategy is formulated for better matching with the demand of market orders, and the cigarette product sales volume trend prediction is a prerequisite condition for formulating the product release strategy. In order to make the product release strategy of the next period, the product sales of the next period must be firstly subjected to predictive analysis. Therefore, how to accurately predict the sales of the cigarette products has important significance for making the putting strategy of the cigarette products.
Disclosure of Invention
In view of this, the embodiment of the invention provides a method and a device for predicting the sales of cigarette products based on a deep neural network, so as to solve the problem that the sales of cigarette products are difficult to predict accurately in the prior art.
According to a first aspect, an embodiment of the present invention provides a method for predicting cigarette product sales based on a deep neural network, including:
step S1: the method comprises the following steps of constructing a neural network model for predicting the sales volume of the cigarette products, wherein an input neuron of the neural network model is composed of a machine extraction characteristic unit and an expert extraction characteristic unit, the machine characteristic extraction unit comprises the sales volume time sequence characteristic of the cigarette products extracted from historical sales data, the expert extraction unit is composed of characteristic data selected by experts according to the historical sales data, and the characteristic data comprise: the sales statistical characteristic and the nonlinear characteristic of cigarette products, and the civil data characteristic and the holiday information characteristic related to the cigarette sales, wherein each characteristic of the machine extraction characteristic unit and the expert extraction characteristic unit is used as one input neuron;
step S2: acquiring periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values in the machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in the expert extraction unit according to the periodic historical sales data;
step S3: according to a fixed preset time sequence period, correspondingly inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model for model training to obtain a sales prediction value, performing error feedback according to the relationship between the sales prediction value and a real sales value in the historical sales data, correcting the weight of each input neuron, then, according to the fixed preset time sequence period again, correspondingly inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model for model training until all training samples are trained, and obtaining a cigarette product sales prediction model;
step S4: and acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales statistical characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current holiday information characteristic parameter value into the cigarette product sales prediction model to acquire a sales prediction value of the cigarette product in the next preset time sequence period.
Optionally, the step S3 specifically includes:
step S31: acquiring all characteristic parameter values corresponding to historical cigarette product sales data of a current preset time sequence period, inputting the characteristic parameter values serving as training samples into input neurons of the neural network model for model training to obtain a sales prediction value of a next preset time sequence period, and obtaining the next preset time sequence period by taking one week as a time sequence sliding window on the basis of the current time sequence period;
step S32: calculating an error between the predicted sales value of the next preset time sequence period and the real sales value of the next preset time sequence period, inputting the error into the neural network model for reverse calculation, and correcting the weight of each input neuron according to a calculation result;
step S33: and taking the next preset time sequence period as the current preset time sequence period, and returning to the step S31 until all training samples are trained, so as to obtain the cigarette product sales prediction model.
Optionally, the neural network model is a fully-connected BP neural network and includes an activation layer, and the output neurons of the neural network model are predicted sales values of cigarette products.
Optionally, the machine extraction feature unit is composed of a long-short term memory deep neural network, the historical sales data in the fixed preset time sequence period is input into the long-short term memory deep neural network, two time sequence feature values are calculated and output, the two time sequence feature values are set as two input neurons of the neural network model, and the two time sequence feature values are input into the neural network model for model training.
Optionally, the preset time sequence period is four weeks, or six weeks, or eight weeks.
Optionally, the cigarette product sales statistical characteristics at least include: the maximum value, the minimum value, the average value, the intermediate value, the mean square error, the variation coefficient and the root mean square of the sales; the non-linear characteristics of the smoking article comprise at least: first order skewness, second order skewness, curvature, KL divergence.
Optionally, the demographics data feature is demographics data of a cigarette product sales forecasting region; the holiday information is characterized by the holidays that exist each week and the type of the holiday.
According to a second aspect, an embodiment of the present invention provides a device for predicting sales of cigarette products based on a deep neural network, including:
the model building module is used for building a neural network model for predicting the sales volume of the cigarette products, and input neurons of the neural network model are composed of a machine extraction feature unit and an expert extraction feature unit, wherein the machine feature extraction unit extracts the sales volume time sequence features of the cigarette products from historical sales data, the expert extraction unit is composed of feature data selected by experts according to the historical sales data, and the feature data comprise: the sales statistical characteristic and the nonlinear characteristic of cigarette products, and the civil data characteristic and the holiday information characteristic related to the cigarette sales, wherein each characteristic of the machine extraction characteristic unit and the expert extraction characteristic unit is used as one input neuron;
the sample extraction module is used for acquiring periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values in the machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in the expert extraction unit according to the periodic historical sales data;
the model training module is used for inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to a fixed preset time sequence period to perform model training to obtain a sales prediction value, performing error feedback according to the relationship between the sales prediction value and a real sales value in the historical sales data, correcting the weight of each input neuron, then inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to the fixed preset time sequence period again to perform model training until all training samples are trained, and obtaining a cigarette product sales prediction model;
and the sales prediction module is used for acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current festival and holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales time sequence characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current festival and holiday information characteristic parameter value into the cigarette product sales prediction model to obtain a sales prediction value of the next preset time sequence period of the cigarette product.
According to a third aspect, an embodiment of the present invention further provides an electronic device, including: the device comprises a memory and a processor, wherein the memory and the processor are in communication connection with each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method for predicting the sales of the cigarette products based on the deep neural network according to the first aspect of the embodiments of the present invention and any optional implementation manner thereof.
According to a fourth aspect, the embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores computer instructions, where the computer instructions are configured to cause the computer to execute the method for predicting the sales of cigarette products based on the deep neural network according to the first aspect of the embodiment of the present invention and any one of the optional implementation manners of the first aspect of the embodiment of the present invention.
The technical scheme of the invention has the following advantages:
the embodiment of the invention provides a method and a device for predicting the sales volume of cigarette products based on a deep neural network, which comprises the steps of constructing a neural network model for predicting the sales volume of the cigarette products, obtaining a sales volume time sequence characteristic parameter value, a sales volume statistical characteristic parameter value, a nonlinear characteristic parameter value, a civil data characteristic parameter value and a holiday information characteristic parameter value as training data of each input neuron in the neural network model by utilizing periodic historical sales data of the cigarette products, extracting each characteristic parameter value as a training sample according to a fixed preset time sequence period, inputting the training sample into the neural network model for model training to obtain a predicted value of the sales volume of the next preset time sequence period, and then utilizing the relation between the predicted value of the sales volume and a real sales volume value for error feedback to correct the weight of each input neuron until all the training samples are trained to obtain a cigarette product sales volume prediction model, and then obtaining all current characteristic parameter values from the actual sales data of the current time sequence period, and inputting the current characteristic parameter values into a cigarette product sales prediction model to predict a sales prediction value of the next preset time sequence period. Therefore, the time sequence characteristics obtained from the historical sales data and the characteristic data selected by experts according to the historical sales data are used as training samples of the neural network model, the time sequence characteristics of the sales data are considered, the sales statistical characteristics and the nonlinear characteristics of the cigarette products obtained by referring to the expert experience, the demographics data characteristics and the holiday information characteristics related to the cigarette sales, the accuracy of the neural network model for predicting the cigarette product sales after training is improved, the prediction result of the trained cigarette product sales prediction model is more fit with the actual sales condition, and the accuracy of the cigarette product sales prediction is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a method for predicting the sales of cigarette products based on a deep neural network in an embodiment of the present invention;
FIG. 2 is a schematic diagram of a model for predicting the sales of cigarette products in an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a cigarette product sales prediction device based on a deep neural network in an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electronic device in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The technical features mentioned in the different embodiments of the invention described below can be combined with each other as long as they do not conflict with each other.
As the monopoly management is carried out in China, in order to respond to the demands of supply side reform related to the industry chain and strengthen the industry layout optimization taking market as the guide, the structures of cigarette product sale and production and manufacture need to be continuously adjusted and optimized, so as to provide cigarette products which better meet the requirements of consumers. The source of the development of the tobacco industry chain is at the consumption end, and a set of scientific and effective cigarette consumption prediction method and cigarette product putting strategy method is beneficial to guiding the market consumption demand and reducing the waste, and realizes the basic situation of 'slightly tight balance' between the supply end and the consumption end.
In addition, the release of cigarette products is a very important basic task, and the sales orders of cigarette products brought by the release of products directly influence the economic benefits of commercial companies. The essence of cigarette release is to carry out sales volume prejudgment and strategic guidance on cigarette sales work at the next stage, the release strategy is formulated for better matching with the demand of market orders, and the cigarette product sales volume trend prediction is a prerequisite condition for formulating the product release strategy. Therefore, in order to make a product release strategy for the next cycle, a predictive analysis must first be performed on the product sales volume for the next cycle.
Based on the above, the embodiment of the present invention provides a method for constructing a prediction model for predicting the sales of a cigarette product, which is used for constructing a prediction model for predicting the sales of a cigarette product, and as shown in fig. 1, the method for constructing a prediction model for predicting the sales of a cigarette product comprises the following steps:
step S1: the method comprises the following steps of constructing a neural network model for predicting the sales volume of the cigarette products, wherein an input neuron of the neural network model is composed of a machine extraction characteristic unit and an expert extraction characteristic unit, the machine characteristic extraction unit comprises the sales volume time sequence characteristic of the cigarette products extracted from historical sales data, the expert extraction unit is composed of characteristic data selected by experts according to the historical sales data, and the characteristic data comprise: the method comprises the steps of obtaining sales statistical characteristics and nonlinear characteristics of cigarette products, and civil data characteristics and holiday information characteristics related to cigarette sales, and using each characteristic of a machine extraction characteristic unit and an expert extraction characteristic unit as an input neuron.
Specifically, in an embodiment, the neural network model adopts a fully-connected BP neural network, the fully-connected BP neural network comprises an activation layer, and output neurons of the neural network model are predicted sales values of cigarette products.
Specifically, in an embodiment, the machine extraction feature unit is formed by a long-short term memory deep neural network, and the historical sales data in a fixed preset time sequence period is input into the long-short term memory deep neural network, two time sequence feature values are calculated and output, the two time sequence feature values are set as two input neurons of the neural network model, and the two time sequence feature values are input into the neural network model for model training.
Specifically, in an embodiment, the statistical characteristics of the sales amount of the cigarette product include: maximum value of the sales, minimum value of the sales, average value of the sales, median value of the sales, mean square error of the sales, coefficient of variation of the sales, root mean square of the sales, etc.; the non-linear characteristics include: first order skewness, second order skewness, curvature, KL divergence and the like; the civil data characteristics related to the sales of the cigarette products comprise: features extracted from regional livelihood data downloaded from a national statistics website; the holiday information characteristics include: holidays and their types (legal holidays, special holidays, traditional holidays), the holidays and their types existing in each week can be taken as a holiday information feature in practical application.
And inputting all characteristic values into the neural network model for deduction calculation by taking each characteristic as an input neuron in the neural network model. Because the characteristics obviously influence the sales volume in the actual sales process of the cigarette products, the input characteristics of model training on the neural network are richer by fully utilizing the characteristics, so that the accuracy of the final cigarette product sales volume prediction model is improved, and the sales volume result predicted by the cigarette product sales volume prediction model is more fit with the actual sales condition.
Step S2: the method comprises the steps of obtaining periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values in a machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in an expert extraction unit according to the periodic historical sales data.
Step S3: and according to a fixed preset time sequence period, all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit are used as training samples and correspondingly input into each input neuron of the neural network model for model training to obtain a sales predicted value, error feedback is carried out according to the relation between the sales predicted value and a real sales value in historical sales data, the weight of each input neuron is corrected, then according to the fixed preset time sequence period, all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit are used as training samples and correspondingly input into each input neuron of the neural network model for model training, and until all training samples are trained, the cigarette product sales prediction model is obtained. In practical applications, the fixed preset time sequence period may be set according to actual requirements, and may be set to four weeks, six weeks, or eight weeks in order to meet the actual rules of the sales volume of the cigarette products, which is not limited herein.
Specifically, in an embodiment, the step S3 includes the following steps:
step S31: and acquiring all characteristic parameter values corresponding to the historical sales data of the cigarette products in the current preset time sequence period, inputting the characteristic parameter values serving as training samples into each input neuron of the neural network model for model training to obtain a sales prediction value of the next preset time sequence period, and taking one week as a time sequence sliding window on the basis of the current time sequence period to obtain the next preset time sequence period.
Step S32: and calculating the error between the predicted sales value of the next preset time sequence period and the real sales value of the next preset time sequence period, inputting the error into the neural network model for reverse calculation, and correcting the weight of each input neuron according to the calculation result.
Step S33: and taking the next preset time sequence period as the current preset time sequence period, and returning to the step S31 until all training samples are trained, so as to obtain the cigarette product sales prediction model.
Specifically, the process principle of training the neural network model in step S3 is as follows: taking every 8 weeks as a fixed preset time sequence period, taking every 1 week as a time sequence sliding window, taking all characteristic parameter values in a machine characteristic extraction unit and an expert extraction unit within continuous 8 weeks (assumed as 1 week to 8 weeks in the historical sales data) extracted from the historical sales data as a training sample, inputting the training sample into each corresponding input neuron in the constructed neural network model, outputting a predicted sales value of the next continuous 8 weeks (namely 2 nd to 9 th weeks) by a neural network model output neuron, then comparing the predicted sales value with a real sales value of the 2 nd to 9 th weeks, inputting an error into the neural network model for reverse calculation to correct the weight of each input neuron in the neural network model, and then taking all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit corresponding to the 2 nd to 9 th weeks as a new training sample, and training the neural network model, correcting the weight of each input neuron of the neural network model again, and repeating the steps until all training samples consisting of the characteristic data extracted from the historical sales data are trained, and finally obtaining a stable neural network model for the cigarette product sales quantity time sequence prediction, namely the cigarette product sales quantity prediction model.
So that by performing the cyclic extraction of all the above characteristic parameter data with a time-series sliding window of every 1 week, namely, the characteristic parameter data obtained each time has larger data overlap with the previous characteristic parameter data, the training samples of the model are enriched, the problem that the deep neural network model training data samples are not sufficient is effectively solved, and the characteristic parameter data are used as training samples to train the neural network model, and the error feedback and the weight adjustment of each neuron in the neural network model are automatically completed, the difference of the characteristic parameter data between the adjacent preset time sequence periods is smaller, so that the parameter adjustment of the training model is more sensitive, and the accuracy of the final cigarette product sales prediction model is further improved, so that the sales result predicted by the cigarette product sales prediction model accords with the actual sales condition.
Step S4: and acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales statistical characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current holiday information characteristic parameter value into a cigarette product sales prediction model to obtain a sales prediction value of the cigarette product in the next preset time sequence period.
Through the steps S1 to S4, the cigarette product sales prediction method based on the deep neural network provided by the embodiment of the invention obtains the sales time sequence characteristic parameter value, the sales statistical characteristic parameter value, the nonlinear characteristic parameter value, the civil data characteristic parameter value and the holiday information characteristic parameter value as the training data of each input neuron in the neural network model by constructing the neural network model for cigarette product sales prediction and utilizing the periodic historical sales data of the cigarette product, extracts each characteristic parameter value as a training sample to input into the neural network model for model training according to a fixed preset time sequence period to obtain the sales prediction value of the next preset time sequence period, and utilizes the relationship between the sales prediction value and the real sales value to carry out error feedback to correct the weight of each input neuron until all training samples are trained, and obtaining a cigarette product sales prediction model, and then obtaining all current characteristic parameter values from the actual sales data of the current time sequence period, inputting all the current characteristic parameter values into the cigarette product sales prediction model to predict the sales prediction value of the next preset time sequence period. Therefore, the time sequence characteristics obtained from the historical sales data and the characteristic data selected by experts according to the historical sales data are used as training samples of the neural network model, the time sequence characteristics of the sales data are considered, the sales statistical characteristics and the nonlinear characteristics of the cigarette products obtained by referring to the expert experience, the demographics data characteristics and the holiday information characteristics related to the cigarette sales, the accuracy of the neural network model for predicting the cigarette product sales after training is improved, the prediction result of the trained cigarette product sales prediction model is more fit with the actual sales condition, and the accuracy of the cigarette product sales prediction is improved.
The method for constructing the prediction model of the cigarette product sales provided by the embodiment of the invention will be described in detail below with reference to specific application examples.
The method for constructing the prediction model of the cigarette product sales volume provided by the embodiment of the invention adopts a multi-layer neural network formed by a full-connection BP neural network and an LSTM long-short term memory neural network to construct the neural network model of the cigarette product sales volume prediction, and takes the data characteristics (including all the characteristics) selected from historical sales order data as the input neurons in the full-connection BP neural network, and the output neurons are the predicted sales volume values, as shown in FIG. 2. The feature extraction is a key step for constructing a neural network model and can be specifically divided into machine feature extraction and expert feature extraction. The operation principle of the whole neural network model is that every 4 weeks, 6 weeks and 8 weeks are used as a time sequence, every 1 week is a time sequence sliding window, historical data in continuous 8 weeks (assumed to be the 1 st to 8 th weeks) are used as training samples and input into the neural network model for calculation, an output predicted value is compared with an actual sales value, errors are calculated in a reverse direction, the weight of each neuron in the neural network model is corrected, then historical data in the next 8 weeks (the 2 nd to 9 th weeks) are input, the weight is corrected again, and the like until the training data samples are finished, and finally, the stable neural network model for predicting the sales volume of cigarette products is obtained.
The construction of the prediction model and the use of the prediction model as shown in fig. 2 are as follows:
s101, constructing a cigarette product sales prediction model formed by a neural network, wherein an input neuron of the model is formed by a machine extraction characteristic unit and an expert extraction characteristic unit 2; the machine extraction characteristic unit is composed of a long-term and short-term memory deep neural network (LSTM), and time sequence characteristics of the cigarette product sales volume are captured by inputting historical sales order data; the expert extraction characteristic unit is formed by characteristic data selected by experts and comprises cigarette product sales statistical characteristics and nonlinear characteristics, and civil data characteristics and holiday information characteristics related to cigarette sales;
step S102, taking periodic historical sales data of cigarette products as input, and respectively calculating to obtain sales quantity time sequence characteristic parameter values in a machine extraction characteristic unit and non-time sequence characteristic parameter values in an expert extraction characteristic unit;
step S103, inputting characteristic parameter values in a machine extraction characteristic unit and an expert extraction characteristic unit into a neural network for model training in a fixed time sequence period, wherein the neural network model can automatically complete error feedback and neuron weight adjustment in the training process;
and step S104, calculating to obtain a time sequence characteristic parameter value and a non-time sequence characteristic parameter value of the sales by taking the actual sales data of the period as input, inputting the characteristic values into the trained prediction model, and calculating to obtain a sales predicted value of the cigarette product in the next period.
The embodiment of the present invention further provides a device for predicting the sales of cigarette products based on a deep neural network, as shown in fig. 3, the device for predicting the sales of cigarette products based on a deep neural network includes:
the model building module 1 is used for building a neural network model for predicting the sales volume of cigarette products, an input neuron of the neural network model is composed of a machine extraction characteristic unit and an expert extraction characteristic unit, wherein the machine characteristic extraction unit comprises the sales volume time sequence characteristic of the cigarette products extracted from historical sales data, the expert extraction unit is composed of characteristic data selected by experts according to the historical sales data, and the characteristic data comprises: the method comprises the steps of obtaining sales statistical characteristics and nonlinear characteristics of cigarette products, and civil data characteristics and holiday information characteristics related to cigarette sales, and using each characteristic of a machine extraction characteristic unit and an expert extraction characteristic unit as an input neuron. For details, refer to the related description of step S1 in the above method embodiment, and no further description is provided here.
And the sample extraction module 2 is used for acquiring periodic historical sales data of the cigarette products, and obtaining sales quantity time sequence characteristic parameter values in the machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in the expert extraction unit according to the periodic historical sales data. For details, refer to the related description of step S2 in the above method embodiment, and no further description is provided here.
And the model training module 3 is used for inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to a fixed preset time sequence period to perform model training to obtain a sales predicted value, performing error feedback according to the relation between the sales predicted value and a real sales value in historical sales data, correcting the weight of each input neuron, then inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to the fixed preset time sequence period again to perform model training until all training samples are trained, and obtaining the cigarette product sales prediction model. For details, refer to the related description of step S3 in the above method embodiment, and no further description is provided here.
And the sales prediction module 4 is used for acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales time sequence characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current holiday information characteristic parameter value into the cigarette product sales prediction model to obtain a sales prediction value of the cigarette product in the next preset time sequence period. For details, refer to the related description of step S4 in the above method embodiment, and no further description is provided here.
Through the cooperative cooperation of the above components, the cigarette product sales prediction device based on the deep neural network provided by the embodiment of the invention obtains the sales time sequence characteristic parameter value, the sales statistical characteristic parameter value, the nonlinear characteristic parameter value, the civil data characteristic parameter value and the holiday information characteristic parameter value as the training data of each input neuron in the neural network model by constructing the neural network model for the cigarette product sales prediction and utilizing the periodic historical sales data of the cigarette product, extracts each characteristic parameter value as the training sample to input into the neural network model for model training according to the fixed preset time sequence period to obtain the sales prediction value of the next preset time sequence period, and then utilizes the relationship between the sales prediction value and the real sales value to carry out error feedback to correct the weight of each input neuron until all the training samples are trained, and obtaining a cigarette product sales prediction model, and then obtaining all current characteristic parameter values from the actual sales data of the current time sequence period, inputting all the current characteristic parameter values into the cigarette product sales prediction model to predict the sales prediction value of the next preset time sequence period. Therefore, the time sequence characteristics obtained from the historical sales data and the characteristic data selected by experts according to the historical sales data are used as training samples of the neural network model, the time sequence characteristics of the sales data are considered, the sales statistical characteristics and the nonlinear characteristics of the cigarette products obtained by referring to the expert experience, the demographics data characteristics and the holiday information characteristics related to the cigarette sales, the accuracy of the neural network model for predicting the cigarette product sales after training is improved, the prediction result of the trained cigarette product sales prediction model is more fit with the actual sales condition, and the accuracy of the cigarette product sales prediction is improved.
There is also provided an electronic device according to an embodiment of the present invention, as shown in fig. 4, the electronic device may include a processor 901 and a memory 902, where the processor 901 and the memory 902 may be connected by a bus or in another manner, and fig. 4 takes the example of being connected by a bus as an example.
The memory 902, which is a non-transitory computer readable storage medium, may be used for storing non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the methods in the method embodiments of the present invention. The processor 901 executes various functional applications and data processing of the processor by executing non-transitory software programs, instructions and modules stored in the memory 902, that is, implements the methods in the above-described method embodiments.
The memory 902 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required for at least one function; the storage data area may store data created by the processor 901, and the like. Further, the memory 902 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 902 may optionally include memory located remotely from the processor 901, which may be connected to the processor 901 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
One or more modules are stored in the memory 902, which when executed by the processor 901 performs the methods in the above-described method embodiments.
The specific details of the electronic device may be understood by referring to the corresponding related descriptions and effects in the above method embodiments, and are not described herein again.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware related to instructions of a computer program, and the program can be stored in a computer readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk (Hard Disk Drive, abbreviated as HDD) or a Solid State Drive (SSD), etc.; the storage medium may also comprise a combination of memories of the kind described above.
Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.
Claims (10)
1. A cigarette product sales prediction method based on a deep neural network is characterized by comprising the following steps:
step S1: the method comprises the following steps of constructing a neural network model for predicting the sales volume of the cigarette products, wherein an input neuron of the neural network model is composed of a machine extraction characteristic unit and an expert extraction characteristic unit, the machine characteristic extraction unit comprises the sales volume time sequence characteristic of the cigarette products extracted from historical sales data, the expert extraction unit is composed of characteristic data selected by experts according to the historical sales data, and the characteristic data comprise: the sales statistical characteristic and the nonlinear characteristic of cigarette products, and the civil data characteristic and the holiday information characteristic related to the cigarette sales, wherein each characteristic of the machine extraction characteristic unit and the expert extraction characteristic unit is used as one input neuron;
step S2: acquiring periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values in the machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in the expert extraction unit according to the periodic historical sales data;
step S3: according to a fixed preset time sequence period, correspondingly inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model for model training to obtain a sales prediction value, performing error feedback according to the relationship between the sales prediction value and a real sales value in the historical sales data, correcting the weight of each input neuron, then, according to the fixed preset time sequence period again, correspondingly inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model for model training until all training samples are trained, and obtaining a cigarette product sales prediction model;
step S4: and acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales statistical characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current holiday information characteristic parameter value into the cigarette product sales prediction model to acquire a sales prediction value of the cigarette product in the next preset time sequence period.
2. The method for predicting the sales of cigarette products based on the deep neural network of claim 1, wherein the step S3 specifically comprises:
step S31: acquiring all characteristic parameter values corresponding to historical cigarette product sales data of a current preset time sequence period, inputting the characteristic parameter values serving as training samples into input neurons of the neural network model for model training to obtain a sales prediction value of a next preset time sequence period, and obtaining the next preset time sequence period by taking one week as a time sequence sliding window on the basis of the current time sequence period;
step S32: calculating an error between the predicted sales value of the next preset time sequence period and the real sales value of the next preset time sequence period, inputting the error into the neural network model for reverse calculation, and correcting the weight of each input neuron according to a calculation result;
step S33: and taking the next preset time sequence period as the current preset time sequence period, and returning to the step S31 until all training samples are trained, so as to obtain the cigarette product sales prediction model.
3. The method for predicting the sales of the cigarette products based on the deep neural network according to claim 1, wherein the neural network model is a fully-connected BP neural network and comprises an activation layer, and output neurons of the neural network model are predicted values of the sales of the cigarette products.
4. The method for predicting the sales of cigarette products based on the deep neural network according to claim 1, wherein the machine extraction feature unit is composed of a long-short term memory deep neural network, historical sales data in the fixed preset time sequence period are input into the long-short term memory deep neural network, two time sequence feature values are calculated and output, the two time sequence feature values are set as two input neurons of the neural network model, and the two time sequence feature values are input into the neural network model for model training.
5. The deep neural network-based cigarette product sales prediction method of claim 1, wherein the preset time sequence period is four weeks, or six weeks, or eight weeks.
6. The method for predicting the sales of cigarette products based on the deep neural network of claim 1, wherein the statistical characteristics of the sales of cigarette products at least comprise: the maximum value, the minimum value, the average value, the intermediate value, the mean square error, the variation coefficient and the root mean square of the sales; the non-linear characteristics of the smoking article comprise at least: first order skewness, second order skewness, curvature, KL divergence.
7. The method for predicting the sales of cigarette products based on the deep neural network according to claim 1, wherein the demographic data is demographic data of a cigarette product sales prediction region; the holiday information is characterized by the holidays that exist each week and the type of the holiday.
8. A cigarette product sales prediction device based on a deep neural network is characterized by comprising the following components:
the model building module is used for building a neural network model for predicting the sales volume of the cigarette products, and input neurons of the neural network model are composed of a machine extraction feature unit and an expert extraction feature unit, wherein the machine feature extraction unit extracts the sales volume time sequence features of the cigarette products from historical sales data, the expert extraction unit is composed of feature data selected by experts according to the historical sales data, and the feature data comprise: the sales statistical characteristic and the nonlinear characteristic of cigarette products, and the civil data characteristic and the holiday information characteristic related to the cigarette sales, wherein each characteristic of the machine extraction characteristic unit and the expert extraction characteristic unit is used as one input neuron;
the sample extraction module is used for acquiring periodic historical sales data of cigarette products, and obtaining sales quantity time sequence characteristic parameter values in the machine characteristic extraction unit and sales quantity statistical characteristic parameter values, nonlinear characteristic parameter values, civil data characteristic parameter values and holiday information characteristic parameter values in the expert extraction unit according to the periodic historical sales data;
the model training module is used for inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to a fixed preset time sequence period to perform model training to obtain a sales prediction value, performing error feedback according to the relationship between the sales prediction value and a real sales value in the historical sales data, correcting the weight of each input neuron, then inputting all characteristic parameter values in the machine characteristic extraction unit and the expert extraction unit as training samples into each input neuron of the neural network model correspondingly according to the fixed preset time sequence period again to perform model training until all training samples are trained, and obtaining a cigarette product sales prediction model;
and the sales prediction module is used for acquiring actual sales data of the current time sequence period, acquiring a current sales time sequence characteristic parameter value in the machine characteristic extraction unit and a current sales statistical characteristic parameter value, a current nonlinear characteristic parameter value, a current civil data characteristic parameter value and a current festival and holiday information characteristic parameter value in the expert extraction unit according to the actual sales data, and inputting the current sales time sequence characteristic parameter value, the current nonlinear characteristic parameter value, the current civil data characteristic parameter value and the current festival and holiday information characteristic parameter value into the cigarette product sales prediction model to obtain a sales prediction value of the next preset time sequence period of the cigarette product.
9. An electronic device, comprising:
a memory and a processor, wherein the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the method for predicting the sales of cigarette products based on the deep neural network according to any one of claims 1 to 7.
10. A computer-readable storage medium storing computer instructions for causing a computer to execute the method for predicting sales of a cigarette product based on a deep neural network according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010661727.XA CN111932292A (en) | 2020-07-10 | 2020-07-10 | Cigarette product sales prediction method and device based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010661727.XA CN111932292A (en) | 2020-07-10 | 2020-07-10 | Cigarette product sales prediction method and device based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111932292A true CN111932292A (en) | 2020-11-13 |
Family
ID=73312847
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010661727.XA Pending CN111932292A (en) | 2020-07-10 | 2020-07-10 | Cigarette product sales prediction method and device based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111932292A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113888235A (en) * | 2021-10-22 | 2022-01-04 | 创优数字科技(广东)有限公司 | Training method of sales forecasting model, sales forecasting method and related device |
| CN114445155A (en) * | 2022-04-08 | 2022-05-06 | 广东烟草佛山市有限责任公司 | Tobacco big data prediction method and system |
| CN114511822A (en) * | 2022-04-19 | 2022-05-17 | 广州市方连科技有限公司 | Daily sundries sales system capable of predicting sales volume according to monitoring picture |
| CN115081890A (en) * | 2022-06-27 | 2022-09-20 | 湖北中烟工业有限责任公司 | Cigarette formula module quality monitoring system, method and related components |
-
2020
- 2020-07-10 CN CN202010661727.XA patent/CN111932292A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113888235A (en) * | 2021-10-22 | 2022-01-04 | 创优数字科技(广东)有限公司 | Training method of sales forecasting model, sales forecasting method and related device |
| CN114445155A (en) * | 2022-04-08 | 2022-05-06 | 广东烟草佛山市有限责任公司 | Tobacco big data prediction method and system |
| CN114511822A (en) * | 2022-04-19 | 2022-05-17 | 广州市方连科技有限公司 | Daily sundries sales system capable of predicting sales volume according to monitoring picture |
| CN114511822B (en) * | 2022-04-19 | 2022-07-12 | 广州市方连科技有限公司 | Daily sundries sales system capable of predicting sales volume according to monitoring picture |
| CN115081890A (en) * | 2022-06-27 | 2022-09-20 | 湖北中烟工业有限责任公司 | Cigarette formula module quality monitoring system, method and related components |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111932292A (en) | Cigarette product sales prediction method and device based on deep neural network | |
| WO2021000556A1 (en) | Method and system for predicting remaining useful life of industrial equipment, and electronic device | |
| CN110019401B (en) | Method, device, equipment and storage medium for predicting part quantity | |
| CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
| Chatterjee et al. | Stock price prediction using time series, econometric, machine learning, and deep learning models | |
| CN108090788B (en) | Advertisement conversion rate estimation method based on time information integration model | |
| CN103310286A (en) | Product order prediction method and device with time series characteristics | |
| CN111680841B (en) | Short-term load prediction method, system and terminal equipment based on principal component analysis | |
| CN110751318A (en) | IPSO-LSTM-based ultra-short-term power load prediction method | |
| CN113298288A (en) | Power supply station operation and maintenance cost prediction method integrating time sequence and neural network | |
| CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| CN114707772B (en) | Power load prediction method and system based on multi-feature decomposition and fusion | |
| CN116663746A (en) | Power load prediction method and device, computer equipment and storage medium | |
| CN115145901A (en) | Multi-scale-based time series prediction method and system | |
| CN112288140A (en) | Keras-based short-term power load prediction method, storage medium and equipment | |
| CN114971748A (en) | Prediction data generation method, model training method, computer device, and storage medium | |
| Kotillová | Very short-term load forecasting using exponential smoothing and ARIMA models | |
| CN114626898A (en) | Sales forecasting method, tool, system, equipment and storage medium | |
| US20180260902A1 (en) | System and method for determining an optimal stopping time | |
| CN115545362B (en) | New energy medium-term power prediction method combining artificial intelligence and time sequence decomposition | |
| CN116308482A (en) | Order sales prediction method and device based on neural network model and Xgboost | |
| CN116470485A (en) | Time-series processing system, method and computer-readable storage medium | |
| CN110826196A (en) | Industrial equipment operation data processing method and device | |
| CN113723717B (en) | Method, device, equipment and readable storage medium for predicting short-term load before system day | |
| CN114971736A (en) | Power metering material demand prediction method and device, electronic equipment and storage medium |
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 |