Specific embodiment
Specific embodiments of the present invention will be described in detail with reference to the accompanying drawing.
Present embodiment is a kind of electronic cigarette product line planing method for considering the market factor, is included the following steps:
Step 1: determining the electronic cigarette product attribute for considering the market factor;
The market factor includes electronic cigarette type and electronic cigarette product attribute.
Step 1-1: electronic cigarette type is determined;
Taobao, Jingdone district and all kinds of electronic cigarette forums have been investigated, it is found that electronic cigarette available on the market can be divided into four types
Type: smoking cessation electronic cigarette replaces cigarette type electronic cigarette, class cigarette type electronic cigarette and high-grade gift type electronic cigarette.
Step 1-2: comment on commodity is acquired using web crawlers technology;
The comment on commodity for crawling Jingdone district website using crawler technology finally obtains about 240,000 and comments as analysis sample
By.
Step 1-3: comment on commodity is analyzed using participle technique;
Step 1-3-1: participle obtains candidate feature word;
Participle cleaning is carried out to comment, noun of the frequency greater than 20 times is obtained as shown in table 1, table 2, herein, only shows
The Feature Words of two kinds of electronic cigarettes, table 1 indicate that the candidate feature word of smoking cessation electronic cigarette, table 2 indicate to replace the candidate of cigarette type electronic cigarette
Feature Words.
1 candidate feature word (1) of table
2 candidate feature word (2) of table
Step 1-3-2: the different degree of customer demand is calculated;
The calculation method of the different degree of customer demand occurs for the number that certain Feature Words occurs divided by all Feature Words total
Number.
Step 1-4: product feature is converted to product attribute using QFD technology;
QFD is consumer demand to be converted into design requirement, components characteristic, the multilayer of technique requirement, production requirement
Secondary Deductive method method, robust Design and quality assurance for guide product;
According to the feedback of user's use information, 14 customer demands that customer may be concerned about are summed up, table 3 gives
The correlativity matrix of QFD Planning Model for House of Quality, with the correlation between digital representation product feature and product attribute, wherein 0 indicates
Not related, 1 indicates that relationship is not close, and 3 indicate that relationship is general, and 5 indicate close relation, and 2 and 4 respectively indicate in 1-3 and 3-5
Between relationship;
Relationship between 3 customer demand of table and engineering characteristic
By using Matrix Analysis Method, the weight of product feature is converted into the various products to meet customer demand
The weight of attribute, thus the product attribute for obtaining smoking cessation electronic cigarette and needing to plan for cigarette type electronic cigarette, to each product attribute
After having divided detailed attribute level, the specifying information as shown in table 4, table 5 is obtained.Wherein, table 4 gives smoking cessation electronics
The product attribute and attribute level of cigarette, table 5 give for the product attribute and attribute level of cigarette type electronic cigarette;
By above step, the product attribute of decision is needed to be reduced to 9 from 18, reduces the burden of surveyee,
The accuracy of prediction is improved simultaneously.
The horizontal relationship (1) of each product attribute of 4 electronic cigarette of table
The horizontal relationship (2) of each product attribute of 5 electronic cigarette of table
Step 2: establishing electronic cigarette product line Optimized model;
Assuming that H enterprise will develop new product line to meet the actual demand of different customer subdivisions.Include in the product line
The product of fixed number, and in product the number of the attribute level of each attribute be it is fixed, the attribute level of product attribute is not
Together, cause these product functions similar, slightly distinguished in performance;
It is analyzed by the market survey of early period and customer demand, H company objective product line is corresponded into the customer in market and is gathered
Class is I and segments market, wherein each the segmenting market including niThe similar customer of (i=1,2 ..., I) a preference, meanwhile, city
There is M competing product in field.With stating the attribute level of each product attribute based on the perception utility surplus of amount of money for caring for
The price that the significance level of visitor, i.e. customers' perception effectiveness subtract product variant.Therefore, in the Optimized model towards profit, valence
Lattice are a decision variables.The present invention is using the price of product variant and specific products configuration as main decision variables, to reach
To the maximized purpose of product line gross profit;
Symbol involved in model is described below:
The quantity that I segments market;
The quantity of J product variant;
niIndicate the customer quantity to segment market in i;
PijIn the segmenting market at i-th, a possibility that jth kind product variant is selected;
The quantity of K product attribute;
LkThe horizontal quantity of k-th of product attribute;
CklVariable cost corresponding to first of level of k-th of product attribute;
CfThe fixed cost of product development;
CvThe variable cost of product development;
The totle drilling cost of C product development;
The zoom factor of μM NL rule;
UijIt segments market at i-th, the value of utility of jth kind product;
uiklK-th of product attribute of product, the element utility value that first of attribute level segments market at i-th;
γijIndicate utility surplus value of the product variant j in the i that segments market;
Indicate utility surplus value of the competing product j in the i that segments market;
M competing product quantity;
The overall market profit of G product line;
The overall market of R product line is taken in;
wjkThe weight of k-th of product attribute of j-th of product;
The symbol of decision variable is expressed as:
xjklIt is a 0-1 variable, is 1 as first of attribute level of fruit product variant k-th of attribute of j is selected duration,
It otherwise is 0;
pjThe price of j-th of product variant.
Step 2-1: customers' preferences analysis is carried out using unified analysis method;
It carries out Consumer Preferences and analyzes most common method to be unified analysis method, explore the different product attribute configuration of product
Influence to the purchase decision of customer obtains joint preference structure (Additive Conjoint using statistical analysis technique
Preference Structure).When by the preference of unified analysis method analysis customer, it is necessary first to allow customer to one group of tool
The form for having the virtual product of different attribute horizontal combination to give a mark is assessed, and least square regression, level shellfish are then used
The methods of Ye Si is handled come the evaluation result obtained to these investigation, to estimate the ingredient effect of each product attribute level
With value;
In unified analysis method, there are many kinds of can be in the method for computation attribute horizontal component value of utility, according to most common
Element utility analysis model, the overall utility value of a product variant are considered the linear of various pieces element utility value
Superposition, formula are as follows:
In general, customer depends primarily on customer to the buying behavior of Mr. Yu's product variant and is perceived to the product variant
Utility surplus.The utility surplus (being indicated with currency) of one product variant is the value of utility (being indicated with currency) of the product and its
The difference of selling price, formula are as follows:
γij=Uij-pj, i=1,2 ..., I;J=1,2 ..., J
Step 2-2: rule is selected to analyze customer purchasing behavior using customer MNL;
Consumer is not only influenced the buying behavior of a new product by the utility surplus of product itself, also by oneself
The effectiveness Retained of existing similar product is remaining to be influenced.In recent years, probability selection rule was repeatedly used, because rule is describing thus
It is more true and reliable when customer buying behavior.It is selected in rule in probability customer, most common is exactly multinomial decilog
MNL selection rule, the formula of this rule are as follows:
Wherein, μ is a zoom factor, if the value of μ is very big, this model is similar to Deterministic rules;If the value of μ
It is similar to zero, this model is similar to be uniformly distributed.The value of μ can be corrected by the finding of actual market share.
Step 2-3: product line cost analysis;
Product line cost of the invention is divided into two parts of variable cost and fixed cost, so enterprise product line
Totle drilling cost is as follows:
C=Cf+Cv
Wherein, CvIndicate the variable cost of product line, CfIndicate the fixed cost of product line.The fixed cost packet of product line
Company is included to determine the expense of one product line of construction and determine project initiation expense, administration fee and the capital cost of production product
Deng;In addition, the variable cost of product line includes expense caused by enterprise's production product, such as packing charges, assembly costs, storage
Expense and logistics distribution expense etc..According to the cumulative model (Linear-additive Cost Model) of linear expense and
Variable cost therein is expressed as an attribute level to the percentage contribution of cost, therefore, cost by product utility measure
Function is as follows:
Step 2-4: Optimized model is established;
Step 2-4-1: objective function;
Optimization aim of the invention is product line profit maximization.Wherein, the constant expense of product line and any decision become
Amount is all not related, does not have an impact to the result of objective function, therefore ignores, and following formula is final goal function:
Step 2-4-2: related constraint;
Related necessary constraint is as follows;
(1) for each product attribute in product variant, there is and can only select an attribute level, this constraint can be with
It is expressed as formula:
(2) configuration of any two product variant cannot be just the same in product line.This relationship can be expressed as formula:
Step 2-4-3: Optimized model;
It is final to establish complete non-linear product line Optimized model:
γij=Uij-pj, i=1,2 ..., I;J=1,2 ..., J
xjkl∈{0,1}andpj> 0
Step 3: the intelligent optimization algorithm for the solution that designs a model;
The flow chart of overall algorithm is as shown in Fig. 2, specific step is as follows.
Step 3-1: design chromosome structure and coding mode;
In the chromosome that the present invention designs, item chromosome indicates a product line of Corporation R & D, each chromosome
It is made of J child partition, j-th of child partition includes information relevant to the attribute configuration of j-th of product variant, and every height
Subregion includes K+1 gene, and k-th of gene includes information relevant to k-th of attribute of product, indicates this Attributions selection
Attribute level, the price of the K+1 gene representation product variant, since actual product price is usually an integer,
Carrying out integer coding to the price of product variant can be improved efficiency, and the process that price range is separated into several integers can be described
It is as follows: firstly, analysis consumer survey as a result, obtaining the element utility value of each attribute level of product variant;Then, pass through
The element utility value of each attribute level of product variant and the Price Range of cost estimate product variant;Finally, sliding-model control
The price range acquired obtains W integer.Therefore, indicate that the gene of product variant price can be selected from W integer;
Fig. 3 provides the example of a chromosome coding.There are two product variant in this product line, each product variant from
There are three attributes for consumer demand angle.According to coding rule, the attribute configuration situation of product 1 (child partition 1) is as follows: the 1st
The attribute 1 of a gene representation user demand angle selects the 1st attribute level, the category of the 2nd gene representation user demand angle
Property 2 the 2nd attribute levels of selection, the attribute 3 of the 3rd gene representation user demand angle selects the 3rd attribute level, the 4th
The price of gene representation product 1 is the 6th discretization price.The cataloged procedure of product 2 is similar to product 1, and so there is no need to repeat to introduce.
Step 3-2: initialization of population;
For the diversity for safeguarding population, the chromosome in initial population is randomly generated.The attribute configuration of product is in gene
It is indicated using integer coding, the generation model of the random function of the index value of the attribute level and product price of user demand angle
Enclose is [1, L respectivelyk] and [1, W].
Step 3-3: genetic operator design;
Step 3-3-1: crossover operator design;
Using the method for uniform crossover (Uniform Crossover Method, UCM) in the present invention.Fig. 4 is provided
One chromosome carries out the example of crossover operation.Firstly, binary intersection mask is generated at random, for selecting parent to dye
Whether the corresponding position of body carries out crossover operation, if the value of a certain position of mask is 0, indicates corresponding father's chromosome and female dyeing
The gene position of body swaps;If intersecting the corresponding value of mask is 1, without exchange.Mask in this example is 10101110,
According to the crossover rule introduced above, for father's individual and mother's individual, the attribute level and price rope of the 2nd attribute of product 1
The price index value for drawing value and product 2 needs to swap respectively, and the value of other gene positions of the two chromosomes does not need to hand over
It changes, obtains son individual 1 and son individual 2 as shown in Figure 4.
Step 3-3-2: mutation operator design;
The present invention uses the mode of random variation to select that is, if some chromosome needs to make a variation under given aberration rate
Some gene positions of this chromosome, the value of the gene position is again random in its value range to generate a new legal gene.
Step 3-3-3: chromosome repairing strategy;
There may be identical parts in the chromosome structure after cross and variation operation, i.e., exist in product line
The identical product of products configuration.In actual product line optimization design, the product of same product line should be configuration
Different, so needing to carry out the reparation of chromosome in this case, by infeasible solution feasible robustness.The present invention is made below
The step of chromosome repairing strategy is introduced;
Step 3-3-3-1: judge that each child partition (product) of chromosome (i.e. actual product line) in current population is corresponding
The corresponding genic value (attribute level) of unit (attribute) whether there is identical situation, i.e., inspection identical product line on whether
There are the duplicate product variants of attribute configuration;
Step 3-3-3-2: based on the duplicate individual of existing product attribute configuration, match for duplicate product attribute
It sets, the value of its a certain gene position is made to carry out random variation within the scope of its attribute level, as the legal chromosome after reparation;
Step 3-3-3-3: check in chromosome at this time whether still remain duplicate attribute configuration situation, and if it exists, then
It is transferred to step 3-3-3-2;
Step 3-3-3-4: until repeating aforesaid operations to no duplicate attribute configuration, stop operation.
Step 3-4: fitness function and selection are established;
Step 3-4-1: fitness function;
In the present invention, objective function is as fitness function.
Step 3-4-2: selection;
That select in the present invention is roulette wheel selection (Roulette Wheel Selection).The basic think of of this method
Think are as follows: chromosome is selected to be genetic to follow-on probability to it just when proportional, just when bigger, remains into the next generation by selection
A possibility that it is also bigger.Assuming that the size of population is Popsize, wherein the fitness of individual is fp, then individual p is selected general
Rate PpIt can be expressed as formula:
After obtaining a possibility that each chromosome is selected, for assortative mating individual, needs to carry out take turns more and select,
Uniform random number between one [0,1] is generated in every wheel, and alternatively pointer obtains the individual selected.Selection
Process can be carried out constantly until reaching population scale.
Step 3-5: setting stopping criterion;
The present invention defines stopping criterion using the mode of setting maximum number of iterations, i.e., the number of iterations, which reaches, presets
A threshold value when, search process stop, exporting the optimal solution in current population as final result.
Step 3-6: the genetic algorithm of building insertion hill-climbing algorithm;
The office of traditional genetic algorithm searches that ability is weaker, simultaneously because the complexity of electronic cigarette product line Optimized model, such as
Fruit considers that local search algorithm, the control that bring time consumption also should be stringent is added, this requires that the part introduced is calculated
Method is simple and efficient.Based on this, hill-climbing algorithm is adopted the local search ability for innovatory algorithm.The basic thought of this algorithm is
Search process is compared to hill climbing process, in the case where no any other information about mountain top, along the increased side of height
To going up a hill.The main flow of the algorithm is as follows:
Step 3-6-1: determining primitive solution, is labeled as g;
Step 3-6-2: the position pos=0 that local search starts on initialization chromosome indicates the base that local search starts
Because segment subscript indexes;
Step 3-6-3: if not meeting stopping criterion, i.e., the value of pos is transferred to step 3-6- less than the length of chromosome
4, otherwise, terminate hill-climbing algorithm;
Step 3-6-4: on the current position pos, carrying out local search, even next=NEIGHBOR (g, pos),
Next indicates neighborhood solution at this time;
Step 3-6-5: if the solution performance of the solution performance ratio g of next is more preferable, g=next is enabled, and enable pos=
Pos+1 returns to step 3-6-3, continues local neighborhood search;Otherwise, pos=pos+1 is enabled, step 3-6-3 is returned to.
It will be verified below with a case.
Experimental analysis is carried out using the case of H business-electronic cigarette, which develops two electronic cigarette product lines and come
Meet different the needs of the segmenting market customer to electronic cigarette, the product of every product line production is both positioned at slight smoker and severe
The two segment market smoker, estimate the customer quantity that the two segment market by customer's number of reviews of electric business website and distinguish
For 170000 and 40000.First product line is used to produce the electronic cigarette of smoking cessation, and Article 2 product line replaces cigarette type for producing
Electronic cigarette.The number of all properties level is fixed in product, and the attribute level of product attribute selection is different, leads to these
Electronic cigarette product function is similar, slightly distinguishes in performance, and enterprise needs that suitable attribute level is selected to construct suitable product
Product line is formed, that is, the attribute configuration of product is determined, to reach the strategic purpose of company.
Since the planing method of two product lines is identical, only introduce for each step of cigarette type electronic cigarette product line and specific
Implementation method.
The first step provides the attribute and attribute level of product.It has been given in Table 5 the product attribute for cigarette type electronic cigarette
Attribute level is followed successively by 3,3,2,4,2,2,4,4,3 levels.
All attributes of product and attribute level are combined by second step using orthogonal design, generate virtual product collection.
Since any combination of each product attribute and attribute level is all effective, it is possible that scheme share 3*3*2*4*2*2*4*
4*3=13824.Since the possible products scheme order of magnitude is too big, carried out by IBM SPSS Statistics software orthogonal
Design, obtains 32 representative virtual products, significantly reduces the difficulty of market survey.This 32 virtual products
Attribute configuration information is as shown in table 6 (integer of every lattice represents selected horizontal serial number in table).
The orthographic virtual product of 6 electronic cigarette of table
Third step calculates the value of utility of product attribute.Data are collected, customer is obtained to virtual product by market survey
Evaluation information excavates customer to the preference data of virtual product.Before this experiment expansion, a large amount of customer is invited to carry out market
Investigation collects customer to the preference information of electronics smoke product.Wherein, every interviewee needs virtually to produce 32 of orthogonal design
The effectiveness of product is assessed, and by designing least-squares linear regression method, obtains the element utility value of each attribute level.Together
When, it is assessed by cost of the expert to product, provides variable cost corresponding to the fixed cost and each attribute level of product.
The value of utility and cost of each attribute are as shown in table 7.
The value of utility and cost of each attribute of table 7
4th step calculates the utility surplus value of competing product.Already existing three sections of competing products, specifically match in market
It sets and price is as shown in table 8.It, can in conjunction with the products configuration and price for having competing product in the market on the basis of third step
To obtain the utility surplus value of rival firms' product, i.e., the difference of calculated element utility value and price, calculated result such as table 9
It is shown.
The concrete configuration and price of 8 competing product of table
The utility surplus of 9 competing product of table
5th step estimates the discrete price set of product variant.It is whole for a product variant in abbreviation step 2-1
Body value of utility formula, it is assumed that the weight variable w in formula is set as 1.The element utility value obtained according to table 7 can be estimated every
A product segment market at two in best expectation value of utility:
It can be obtained by result, the best expectation value of utility 179.6 of product variant in two that segments market is in all segment market
Best expectation value of utility, using the value as the upper limit of product variant price.According to the cost of attribute each in table 7 it can be concluded that producing
The highest component purchase cost of product variant:
Highest purchase cost=C13+C23+C31+C44+C52+C61+C74+C84+C93=162
The product variant highest component purchase cost 162 being calculated as the lower limit of product variant price.Therefore, it produces
The price of product variant may be designated as the successive value between 162 and 179.6.Since price is typically all discrete whole in real life
Numerical value, therefore the price continuum of product variant should be carried out sliding-model control, the discrete price set acquired are as follows:
{162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179}。
This experiment is randomly derived initial solution, and enchancement factor influences the observation of the final variation tendency of experimental result in order to prevent,
Every group of experiment can execute several times, and each initial solution is to regenerate.Take the average value of many experiments result as knot
Fruit.
It for problematic features, is solved with genetic algorithm and tabu search algorithm, and carries out the calculation of both algorithms
The optimizing ability comparative experiments of method Experiment Parameter and two kinds of algorithms:
Experiment 1: it is directed to genetic algorithm, the variation of analysis population scale, crossing-over rate variation and aberration rate variation are to optimal
The influence of the average value of solution.
Experiment 2: the comparison of the genetic algorithm and the genetic algorithm for being not embedded into hill-climbing algorithm of hill-climbing algorithm, optimizing ability are embedded in
Comparison.
Experiment 3: it is directed to tabu search algorithm, analysis taboo list size variation and the every time variation of search neighbours' number are to optimal
The influence of the average value of solution.
Experiment 4: the comparison of the genetic algorithm optimizing effect of tabu search algorithm and insertion hill-climbing algorithm.
This experiment compares tabu search algorithm using the example of a small-scale problem and the heredity for being embedded in hill-climbing algorithm
Algorithm, experiment relevant parameter are designed as;The quantity I that segments market is set as 2, and product variant quantity J is set as 2, and zoom factor μ is set as
0.1.The design of algorithm relevant parameter are as follows:
(1) genetic algorithm;Population scale is set as 300, and crossing-over rate is set as 0.85, and aberration rate is set as 0.02, and the number of iterations is set
It is 200 times.
(2) tabu search algorithm: taboo list is sized to 11, searches for neighbours' number every time and is set as 50, the number of iterations is set as
200 times.
Numerical experiment results such as Fig. 5, shown in Fig. 6.It can be learnt by the observation to result: in identical runing time,
The optimizing ability of the genetic algorithm of hill-climbing algorithm is embedded in far better than common genetic algorithm.Meanwhile in identical runing time, lose
The optimizing ability of propagation algorithm is far better than tabu search algorithm.