CN113240359B - Demand prediction method for coping with external major changes - Google Patents
Demand prediction method for coping with external major changes Download PDFInfo
- Publication number
- CN113240359B CN113240359B CN202110341015.4A CN202110341015A CN113240359B CN 113240359 B CN113240359 B CN 113240359B CN 202110341015 A CN202110341015 A CN 202110341015A CN 113240359 B CN113240359 B CN 113240359B
- Authority
- CN
- China
- Prior art keywords
- data
- demand
- prediction
- training
- characteristic data
- 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 47
- 230000010485 coping Effects 0.000 title claims abstract description 10
- 238000012549 training Methods 0.000 claims description 47
- 238000012795 verification Methods 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 18
- 210000002569 neuron Anatomy 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 12
- 238000011144 upstream manufacturing Methods 0.000 claims description 11
- 239000000047 product Substances 0.000 claims description 8
- 238000012706 support-vector machine Methods 0.000 claims description 8
- 238000012417 linear regression Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 230000009977 dual effect Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 239000012467 final product Substances 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000012384 transportation and delivery Methods 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 2
- 230000004044 response Effects 0.000 claims 3
- 238000010801 machine learning Methods 0.000 abstract description 6
- 230000008859 change Effects 0.000 description 8
- 238000007726 management method Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000007613 environmental effect Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000013519 translation Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000001502 supplementing effect Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 229920002430 Fibre-reinforced plastic Polymers 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000011151 fibre-reinforced plastic Substances 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
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
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/087—Inventory or stock management, e.g. order filling, procurement or balancing against orders
- G06Q10/0875—Itemisation or classification of parts, supplies or services, e.g. bill of materials
-
- 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
- G06F16/25—Integrating or interfacing systems involving database management systems
- G06F16/252—Integrating or interfacing systems involving database management systems between a Database Management System and a front-end application
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- General Engineering & Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Operations Research (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Tourism & Hospitality (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Quality & Reliability (AREA)
- Human Resources & Organizations (AREA)
- Fuzzy Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a demand prediction method for coping with external major fluctuation, which fully utilizes abundant big data resources of logistics enterprises, considers the influence of major fluctuation of external environment, utilizes a method combining a plurality of efficient and accurate machine learning models to play the advantages of different models in demand prediction, captures potential factors of the major fluctuation of the external environment by utilizing the machine learning models and the big data, can effectively relieve the influence of the large uncertainty of the prediction target and the external environment on the prediction result, obviously improves the accuracy of the logistics enterprises on demand quantity prediction, and assists in optimizing the decision result of the enterprises.
Description
Technical Field
The invention relates to the technical field of inventory management and prediction, in particular to a demand prediction method for coping with serious external changes.
Background
In recent years, the retail sales of the social commodities in China always keep an ever-increasing situation. The online platform online shopping channel becomes a main way for consumers to purchase products. With the expansion of the online shopping market scale, higher-level requirements are put on the supply chain. In the supply chain, logistics enterprises need to both support the supply of upstream suppliers and to be responsible for distributing products downstream, with a significant impact on supply chain turnaround efficiency.
Regarding to the inventory management flow of the current logistics enterprises, the upstream replenishment and downstream sales conditions need to be predicted, and the prediction is used as one of the basis for making an inventory optimization scheme and an inventory entry and exit plan. The specific problems include: in practical situations, the traditional prediction mode based on experience guidance and manual decision cannot meet the current huge market demand, is difficult to adapt to the upstream and downstream environments of a complex supply chain, and has more defects.
First, the management and decision process data is of low participation. At present, the enterprise information system is gradually popularized and applied, data can be effectively recorded and stored, but when an administrator makes an actual decision, the manager relies on history experience too much, so that data resources cannot be fully utilized to provide auxiliary support for decision making, waste of the data resources is formed, and a great error exists in decision making results.
Secondly, the randomness of the predicted target is larger, and the prediction difficulty is higher. The consumption demand has strong uncertainty and larger variability. Different areas and different time nodes have great difference in commodity demands of different brands and types. In addition, the dramatic change in demand caused by external environmental risks also increases the difficulty of prediction.
Third, significant external environmental changes have a significant negative impact on the accuracy of the predicted results. The external environment has stronger uncertainty, such as epidemic situation and other factors. Such uncertainty often has a severe impact on consumer demand and is difficult to capture by a general forecasting method, so that the quantity of demand forecasted by a logistics enterprise based on a traditional mode is significantly different from the actual quantity.
Fourth, the conventional prediction method is too simple in consideration of factors, and has poor prediction effect. The current prediction method is mainly formed by a prediction method based on historical order data. The method is simple, but the prediction result is too dependent on the experience of the manager, so that the prediction result is often greatly deviated from the actual situation.
In summary, the current logistics enterprise has defects in the demand prediction method, cannot effectively cope with complex prediction situations, and has low accuracy of the prediction results, so that the logistics enterprise is affected to give a reasonable and effective inventory optimization scheme, and the overall collaborative efficiency of the supply chain is low.
Disclosure of Invention
The invention aims to provide a demand prediction method for dealing with serious external changes, which can fully utilize the existing data resources of logistics enterprises, consider the potential change factors of external environments, further effectively improve the prediction result of demand quantity, provide decision support for the enterprise inventory management process, help enterprises to reduce inventory cost and reduce inventory pressure.
The invention aims at realizing the following technical scheme:
a demand prediction method for coping with significant external changes, comprising:
setting the length of a reference history period and demand characteristics, acquiring corresponding demand characteristic data and sales order data in the reference history period as a total data set, dividing the total data set into a training set and a verification set, and preprocessing the demand characteristic data in the training set and the verification set;
respectively training a support vector machine regression model, a continuous limited Boltzmann machine and a gradient lifting tree by using a training set, and obtaining three primary learners after training;
the demand characteristic data in the verification set are respectively input into three primary learners to obtain a prediction result;
using linear regression as a secondary learner, and combining the prediction results to obtain weights corresponding to the three primary learners so as to obtain a linear hybrid total model;
and predicting the product demand quantity by using a linear hybrid total model.
According to the technical scheme provided by the invention, abundant big data resources of logistics enterprises are fully utilized, meanwhile, the influence of great fluctuation of external environment is considered, the advantages of different models in demand prediction are brought into play by means of a method combining a plurality of efficient and accurate machine learning models, meanwhile, potential factors of great fluctuation of external environment are captured by utilizing the machine learning models and big data, the influence of a prediction target and great uncertainty of the external environment on a prediction result can be effectively relieved, the accuracy of the logistics enterprises on demand quantity prediction is remarkably improved, and the decision result of the enterprises is assisted and optimized.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a demand prediction method for coping with serious external changes according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a method for predicting replenishment corresponding to significant external changes according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an overall framework model of a method for replenishment and demand prediction for dealing with significant external changes according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
The embodiment of the invention provides a demand prediction method for dealing with serious external changes, as shown in fig. 1, which mainly comprises the following steps:
step 1, setting the length of a reference history period and demand characteristics, acquiring corresponding demand characteristic data and sales order data in the reference history period as a total data set, dividing the total data set into a training set and a verification set, and preprocessing the demand characteristic data in the training set and the verification set.
The preferred embodiment of this step is as follows:
1) The type of demand feature is set and the reference history period length is set to n.
In the embodiment of the invention, the types of the demand characteristic data mainly comprise: the predicted replenishment quantity, the city of the warehouse, the commodity class, the commodity delivery quantity in the past period (for example, the past five months), the commodity delivery quantity in the past period, the month of the current period and whether the sales promotion is carried out in the current period. The required characteristics are comprehensively analyzed in the training and learning process, the prediction result is continuously calibrated in consideration of the change trend of the warehouse-in and warehouse-out quantity in the past period, and potential external environment risk factors which are difficult to predict in advance by people are timely found out, so that the prediction capable of coping with the important change is made.
2) And acquiring historical n-period sales order data and corresponding type demand feature data, wherein the number of samples in the total data set is n, the demand feature data and corresponding sales order data are contained in a single sample, the demand feature data are vectors formed by all demand features, the vectors are used as input x, and the sales order data are used as demand prediction target real data y.
3) Randomly extracting data with alpha proportion from the total data set as a verification set according to the set fixed data cutting proportion alpha, and marking the data as D val The number of the contained samples is k; the remainder is denoted as D as training set train The number of the contained samples is m, then the numerical data in the demand characteristic data is standardized (namely, the average value and the standard deviation of each demand characteristic in the training set are calculated, then the average value is subtracted from each sample data, the standard deviation is divided by the average value, the data are normalized, and the average value of the training set data is subtracted from the sample data in the verification set, the standard deviation of the training set data is divided by the average value of the training set data), and the type data are subjected to single-heat coding.
Taking the types of the demand characteristic data provided in the foregoing as examples, the final replenishment predicting quantity, the commodity warehouse-out quantity in the past period of time and the commodity warehouse-in quantity in the past period of time are numerical data; the city of the warehouse, the commodity category, the month of the current period and whether the promotion activity is the category type data.
In the embodiment of the invention, the scheme for predicting the replenishment quantity can be realized by adopting the prior art, but the following problems in the prior scheme are considered: 1) Uncertainty in the dual constraint of manufacturer productivity and delivery time affects the logistics enterprise to accurately predict the quantity of supplier replenishment. In addition, the dramatic change in restocking caused by external environmental risks also increases the difficulty of forecasting. 2) Significant external environmental changes have a large negative impact on the accuracy of the predicted results. The external environment has stronger uncertainty, such as epidemic situation and other factors. Such uncertainty often has a severe influence on the replenishment capacity requirement of the supplier, and further causes a large deviation between the replenishment quantity predicted by the logistics enterprises based on the traditional mode and the actual replenishment quantity; however, such uncertainty factors are difficult to capture by a general prediction method, so that a prediction result has a large deviation from an actual situation. Therefore, the embodiment of the invention provides a replenishment prediction scheme capable of coping with serious external changes, the method fully utilizes the existing data resources of a logistics enterprise, considers potential change factors of external environment, and further effectively improves the prediction results of the required quantity, as shown in fig. 2, and is a schematic diagram of the replenishment prediction method capable of coping with serious external changes, which mainly comprises the following steps:
and 11, acquiring the data of the quantity of the replenishment warehouse-in of the upstream supplier according to the set reference historical period length, and preprocessing.
The preferred embodiment of this step is as follows:
1) Setting the length of a reference historical period as n, and acquiring the data of the quantity of the replenishment warehouse-in of the upstream supplier as X (0) ＝ {x (0) (t), t=1, 2, …, n }, where x (0) And (t) supplementing the quantity data of the warehouse entry for the upstream supplier in the t period.
2) Supplementing and warehousing quantity data X for upstream suppliers (0) Performing a level ratio test, and if the level ratio test is not passed, performing translation transformation; the method comprises the following steps:
first calculate the level ratioIf the original data (i.e. data X (0) ) Is satisfied by +.>Then the description data passes the level ratio test and gray prediction can be usedThe model is predicted, otherwise, translation transformation is needed to be carried out on the original data: x is x (0) (t)＝x (0) (t) +c, c is a set value, which enables the level ratio of the original data to satisfy the range.
3) Accumulating the data to obtain preprocessed data X (1) Expressed as
And step 12, obtaining a gray prediction model by using the preprocessed data, and further obtaining preliminary prediction data through the gray prediction model.
The preferred embodiment of this step is as follows:
1) Setting a gray prediction model as GM (1, 1), wherein the specific form is as follows:
x (0) (r)+az (1) (r)＝b
wherein a represents a development coefficient, and b represents a gray action amount; by applying to the preprocessed data X (1) Generating the immediate mean value
2) By obtaining data X of the quantity of the replenishment warehouse of the upstream supplier (0) And the preprocessed data X (1) Obtaining a grey prediction model, comprising:
and solving parameters in the gray prediction model by adopting a least square method, wherein a solving formula is expressed as follows:
wherein,each represents the development coefficient and gray action amount obtained after solving; y is n ＝(x (0) (2),x (0) (3),…,x (0) (n)) T ；/>x (0) (t) is acquired data of the quantity of the replenishment warehouse entry of the upstream supplier in the t-th period, and t=1, 2, …, n and n are set reference historical period lengths.
Gray prediction model parameters estimated by the methodThe predicted value of the gray prediction model is:
by passing throughThe solution on the original data can be further restored from the accumulated data, and the specific formula is as follows: the preliminary prediction data obtained by using the gray prediction model, namely, the preliminary prediction data of the n+1th stage. It should be noted that if the original data passes the level ratio test only by the translation transformation, the data translation c needs to be subtracted from this prediction result.
And 13, setting the number of Markov prediction state intervals and the upper limit and the lower limit of each state interval according to the preliminary prediction data, obtaining a Markov prediction model according to the preliminary prediction data, and further predicting the final product replenishment quantity.
The preferred embodiment of this step is as follows:
1) Setting the number of Markov prediction state intervals and the upper limit and the lower limit of each state interval according to the preliminary prediction data, wherein the method comprises the following steps:
dividing the replenishment quantity into U state sections according to the preliminary prediction data, wherein the U state section is expressed asWherein A is u ，B u Respectively set state intervals E u Upper and lower limits of (2).
Those skilled in the art will appreciate that the upper and lower limits of the state interval are a number, also known as a multiplier, e.g., 0.8,1,1.1. Upper and lower limit values multiplied byThen, a plurality of state sections can be divided. Exemplary, gray prediction model calculated +.>100, A of the first state interval 1 And B 1 0.6 and 0.8, respectively, the first state interval is [60,80 ]]。
2) Obtaining a Markov prediction model according to the preliminary prediction data, wherein the Markov prediction model comprises:
calculating the conditional probability p of the state transition from the ith state interval to the(s) th state interval according to the preliminary prediction data us U, s=1, 2, …, U, where e= { E 1 ,E 2 ,…,E U And is a set of state intervals.
This operation may generate a state transition matrix P of u×u, in the following form:
3) Predicting the final product restocking quantity using the markov prediction model includes:
knowing that the n-th replenishment quantity belongs to the status interval E x X=1, 2, …, U, at conditional probability p xs The largest probability value P is selected among s=1, 2, …, U (i.e. in the x-th row of the state matrix P) xf ＝max s＝1,2,…,U p xs Determining the state interval E of the future n+1th phase f The calculation formula for predicting the final product replenishment quantity is as follows:
wherein A is f And B is connected with f Is a state interval E f Upper and lower limits of (2);the preliminary prediction data is represented, and n is the set reference history period length.
It should be noted that the restocking quantity is mainly used as a demand feature in the demand prediction scheme, and the restocking prediction scheme provided by the embodiment of the invention is used as a preferred scheme and is not the only scheme. On the basis of the demand prediction scheme provided by the invention, even if the quantity of the restocking obtained by adopting the traditional restocking prediction scheme is equal to the quantity of the restocking obtained by adopting the traditional restocking prediction scheme, the final demand prediction effect is still superior to that of the traditional demand prediction scheme.
As shown in fig. 3, the overall frame model is a combination of the demand prediction scheme and the restocking prediction scheme. By inputting the historical restocking data into the restocking prediction method (provided by the invention or a conventional scheme), the restocking quantity predicted by the method can be output. The replenishment quantity can be used as one item of demand characteristic data, and is used as the input of a demand prediction method together with the rest of demand characteristic data and calendar history sales order data, and the predicted demand quantity is output through calculation of the demand prediction method. The inventory management information system is used for accurately predicting the quantity of the restocking and the quantity of the demand, and finally inputting the quantity of the restocking and the quantity of the demand into the existing inventory management information system of the logistics enterprise, so that the inventory management personnel of the logistics enterprise can conveniently check the quantity of the restocking and the quantity of the demand, and the inventory management personnel can be used as an accurate basis for deciding the inventory optimization and the allocation scheme, so that the operation efficiency of the logistics enterprise is improved, the cost is reduced, and meanwhile, the informatization degree and the big data resource utilization efficiency of the logistics enterprise are remarkably improved.
And 2, respectively training a support vector machine for regression by using a training set, and obtaining three primary learners after training by using a continuous limited Boltzmann machine and a gradient lifting tree.
The preferred embodiment of this step is as follows:
1) And training a support vector machine regression model.
Setting a support vector regression model to h 1 (x)＝ω T x+b, so that h 1 (x) As close as possible to y. Where ω and b are parameters to be determined by the model and x represents demand characteristic data. In a support vector machine regression model, h can be tolerated 1 (x) There is a deviation of e at most from y.
The support vector regression problem is formalized as:
introduction of Lagrangian multiplier mu i ≥0,μ i ≥0,α i ≥0,The Lagrangian function is obtained and the conversion of the dual problem is performed, so that the following can be obtained:
in the above formula, alpha,are vectors representing all alpha i ,/>
Solving this dual problem using the KKT condition:
wherein C is regularization constant, ζ i Andis a relaxation variable, alpha i And->Is a lagrange multiplier. Zeta type toy i ≥0,/>
α i ≥0,/>
Solving model parameters and taking feature mapping forms into consideration to obtain model parameters Wherein->y i For demand characteristic data x i Corresponding sales order data, m representing the number of samples in the training set;
finally, obtaining a trained regression model of the support vector machine asWherein, kappa (x) i ,x l ) Is a Gaussian radial basis function, and has the specific form:
(σ>0 is the width of the gaussian kernel).
2) Training a continuous limited boltzmann machine.
Setting a continuous limited boltzmann machine network structure, comprising: the visible layer, the hidden layer and the output layer are sequentially arranged; the visible layer contains the same number of neurons as the number of the demand characteristic data, which is marked as Q, and the output layer contains only one neuron.
Since the conventional boltzmann machine structure can only process binary data, in this embodiment, it is necessary to continuously process the neuron states between the visible layer and the hidden layer and the activation functions thereof on the basis of the conventional boltzmann machine structure. By adding a Gaussian variable with the mean value of 0 and the variance of sigma into neurons, the state of the hidden layer neurons after continuous processing isWherein s is j Representing the state of the jth hidden layer neuron, s q Represent the state of the q-th visible layer neuron, N j (0, 1) represents a random gaussian variable. The activation function of the neuron is rewritten as: /> Activating a variant of a function for sigmoid, where θ H And theta L Respectively representing the upper and lower limits of the neuron state value, a j A parameter indicating the degree of inclination of the sigmoid function to be controlled in the jth neuron.
After the continuous limited Boltzmann machine model is constructed, a contrast divergence algorithm is used and combined with a training set to train network parameters between the visible layer and the hidden layer, weight parameters between the visible layer and the hidden layer are trained, and the bias vector of each neuron is updated.
In the continuous constrained boltzmann machine, neurons in the visible and hidden layers each have their own states. In the contrast divergence algorithm, the initial states of all neurons in the visible layer are randomly initialized, then the states in the hidden layer are obtained in a probability distribution sampling mode, and then the connection weight and the bias vector between the neurons in the two layers are updated in each cycle according to the states of the visible layer and the hidden layer.
Directly outputting the hidden layer and the output layer without activation function conversion, training weight parameters between the hidden layer and the output layer by using a mean square error as a loss function, and obtaining a trained continuous type limited Boltzmann machine h 2 (x) A. The invention relates to a method for producing a fibre-reinforced plastic composite Typically, a maximum number of exercises is preset before the exercises until the maximum number of exercises is reached.
Those skilled in the art will appreciate that the network parameters between the visible layer and the hidden layer, and the weight parameters between the hidden layer and the output layer are two sets of parameters, and that the two sets of parameters are trained in stages. Typically, parameters between the visible layer and the hidden layer are trained first, and then network parameters (i.e., weights) of the hidden layer to the output layer are trained.
3) Training a gradient lifting tree.
Initializing a weak learner by taking the CART regression tree as the weak learner of the gradient lifting tree;
setting training iteration times, taking square loss as a loss function in each iteration, and calculating the negative gradient of the loss function for each demand characteristic data on a training set, wherein the negative gradient of the loss function is a residual error;
the residual error is used as a new true value y of a sample and used for fitting parameter learning to obtain a new CART regression tree (the node area of the regression tree is estimated to fit the approximate value of the residual error), the new CART regression tree is superimposed on the initialized weak learner, and the result is used as an updated strong learner;
repeating the above steps (i.e. continuously adding new CART regression tree to the initial weak learner in each iteration process to serve as updated strong learner) until reaching training iteration times, outputting the strong learner updated in the last iteration as the final learner, wherein the final learner is the gradient lifting tree after training and is marked as h 3 (x)。
Model h obtained by training 1 (x)、h 2 (x)、h 3 (x) As a primary learner.
And step 3, respectively inputting the demand characteristic data in the verification set into three primary learners to obtain a prediction result.
The preferred embodiment of this step is as follows:
for the demand characteristic data in the verification set, three primary learners are respectively used for prediction to obtain the demand characteristic data x in the verification set d The prediction result is denoted as h d ＝(h 1 (x d ),h 2 (x d ),h 3 (x d )) T D=1, 2, …, k, wherein k is the number of samples contained in the verification set, and each sample corresponds to one piece of demand characteristic data; h is a 1 (x d )、h 2 (x d )、h 3 (x d ) Each representing a primary learner for demand characteristic data x d Is a predicted result of (a).
And 4, using linear regression as a secondary learner, and combining the prediction results to obtain weights corresponding to the three primary learners, thereby obtaining a linear hybrid total model.
The preferred embodiment of this step is as follows:
and using linear regression as a secondary learner, and combining the prediction results to obtain weights corresponding to the three primary learners.
The model parameter solving formula of the linear regression is expressed as:
wherein,weights corresponding to three primary learners, < ->For bias item->y＝ (y 1 ,y 2 ,…,y k ) T ，h d For verifying the concentrated demand characteristic data x d And (b) prediction result, y d For demand characteristic data x d Corresponding sales order data, d=1,2, …, k, k is the number of samples contained in the validation set.
And 5, predicting the product demand quantity by using the linear hybrid total model.
The preferred embodiment of this step is as follows:
and taking the future n+1st-stage demand characteristic data as input, respectively predicting by using three primary learners to obtain three primary prediction results, taking the three primary prediction results as input of a linear hybrid total model, fusing the three primary prediction results (namely, calculating the weighted sum of the three primary prediction results) by the linear hybrid total model, and finally predicting the product demand quantity.
According to the method, the quantity of the restocking and the demand in a period of time in the future is accurately predicted according to the needs of the logistics enterprises, the scene of the decision basis of the optimized inventory allocation scheme is used, the abundant big data resources of the logistics enterprises are fully utilized, meanwhile, the influence of the important change of the external environment is considered, the advantages of different models in the forecasting of the restocking and the demand are brought into play by means of a method combining a plurality of efficient and accurate machine learning models, meanwhile, potential factors of the important change of the external environment are captured by utilizing the machine learning models and the big data, the influence of the forecast objective and the large uncertainty of the external environment on the forecast result can be effectively relieved, the accuracy of the logistics enterprises on the forecast of the quantity of the restocking and the quantity of the demand is remarkably improved, and the decision result of the enterprises is assisted to be optimized.
From the above description of embodiments, it will be apparent to those skilled in the art that the above embodiments may be implemented in software or by means of software plus necessary general hardware platforms. Based on such understanding, the technical solution of the foregoing embodiment may be embodied in a software product, where the software product may be stored in a nonvolatile storage medium (may be a CD-ROM, a U-disk, a mobile hard disk, etc.), and includes several instructions for causing a computer device (may be a personal computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.
Claims (9)
1. A demand prediction method for coping with significant external changes, comprising:
setting the length of a reference history period and demand characteristics, acquiring corresponding demand characteristic data and sales order data in the reference history period as a total data set, dividing the total data set into a training set and a verification set, and preprocessing the demand characteristic data in the training set and the verification set; types of the demand features include: the predicted replenishment quantity, the city of the warehouse, the commodity category, the commodity delivery quantity in the past period of time, the month of the current period of time and whether a sales promotion activity exists in the current period of time; the demand characteristic data is a vector composed of all demand characteristics
Respectively training a support vector machine regression model, a continuous limited Boltzmann machine and a gradient lifting tree by using a training set, and obtaining three primary learners after training;
the demand characteristic data in the verification set are respectively input into three primary learners to obtain a prediction result;
using linear regression as a secondary learner, and combining the prediction results to obtain weights corresponding to the three primary learners so as to obtain a linear hybrid total model; the method for obtaining weights corresponding to the three primary learners by combining the prediction results comprises the following steps of: the model parameter solving formula of the linear regression is expressed as: wherein (1)>Weights corresponding to three primary learners, < ->For bias item->y＝(y 1 ,y 2 ,…,y k ) T ，h d For verifying the concentrated demand characteristic data x d And (b) prediction result, y d For demand characteristic data x d Corresponding sales order data, d=1, 2, …, k, k is the number of samples contained in the validation set;
and predicting the product demand quantity by using a linear hybrid total model.
2. The method of claim 1, wherein setting the length of the reference history period and the demand characteristics, and obtaining the corresponding demand characteristics and sales order data in the reference history period as the total data set comprises:
setting the type of the demand characteristic and the length of the reference history period as n;
and acquiring historical n-period sales order data and corresponding type demand feature data, wherein the number of samples in the total data set is n, the demand feature data and corresponding sales order data are contained in a single sample, the demand feature data are vectors formed by all demand features, the vectors are used as input x, and the sales order data are used as demand prediction target real data y.
3. The method of claim 1, wherein the predicting the replenishment amount comprises:
acquiring the data of the quantity of the replenishment warehouse-in of the upstream supplier according to the set reference historical period length, and preprocessing;
obtaining a gray prediction model by using the obtained data of the quantity of the replenishment warehouse-in of the upstream supplier and the preprocessed data, and further obtaining preliminary prediction data by the gray prediction model;
and setting the number of Markov prediction state intervals and the upper limit and the lower limit of each state interval according to the preliminary prediction data, obtaining a Markov prediction model according to the preliminary prediction data, and further predicting the final product replenishment quantity.
4. A method of demand prediction for significant changes in the environment according to claim 1 or 2, wherein the dividing of the total data set into a training set and a validation set, and the preprocessing of the demand characteristic data comprises:
randomly extracting data with alpha proportion from the total data set as a verification set according to the set fixed data cutting proportion alpha, and marking the data as D val The number of the contained samples is k; the remainder is denoted as D as training set train The number of the contained samples is m; and then, carrying out standardization processing on numerical data in the demand characteristic data of the training set and the verification set, and carrying out independent heat coding processing on the category data.
5. A method of demand prediction for significant external changes according to claim 1 or 2, wherein training a support vector machine regression model comprises:
setting a support vector regression model to h 1 (x)＝ω T x+b, ω and b are model parameters to be determined, x represents demand characteristic data;
introducing Lagrangian multiplier to solve support vector regression dual problem to obtain model parameters Where E is the tolerable deviation, α l And->Is Lagrangian multiplier +.>x l For demand characteristic data, y i For demand characteristic data x i Corresponding sales order data, m representing the number of samples in the training set;
finally, obtaining a trained regression model of the support vector machine asWherein, kappa (x) i ,x l ) Is a gaussian radial basis function.
6. A method of demand prediction for response to significant external fluctuations according to claim 1 or 2, wherein the step of training the continuous, constrained boltzmann machine comprises:
setting a continuous limited boltzmann machine network structure, comprising: the visible layer, the hidden layer and the output layer are sequentially arranged; the visible layer comprises the same number of neurons as the required characteristic data, the output layer only comprises one neuron, and the neurons and the activation functions in the visible layer and the hidden layer need to be subjected to continuous processing;
training network parameters between the visible layer and the hidden layer by using a contrast divergence algorithm and combining with a training set;
directly outputting the network parameters between the hidden layer and the output layer without activation function conversion, training the network parameters between the hidden layer and the output layer by using the mean square error as a loss function, and obtaining a trained continuous type limited Boltzmann machine h 2 (x)。
7. A method of demand prediction for response to significant external changes according to claim 1 or 2, wherein the step of training the gradient-lifting tree comprises:
initializing a weak learner by taking the CART regression tree as the weak learner of the gradient lifting tree;
setting training iteration times, taking square loss as a loss function in each iteration, and calculating the negative gradient of the loss function for each demand characteristic data on a training set, wherein the negative gradient of the loss function is a residual error;
taking the residual error as a new true value y of a sample, using the new true value y for fitting parameter learning to obtain a new CART regression tree, superposing the new CART regression tree on the initialized weak learner, and taking the result as an updated strong learner;
repeating training until reaching training iteration times, outputting a strong learner updated in the last step of iteration as a final learner, wherein the final learner is a gradient lifting tree after training, and is marked as h 3 (x)。
8. The method of claim 1, wherein,
the demand characteristic data in the verification set are respectively input into three primary learners, and the obtaining of the prediction result comprises the following steps:
for the demand characteristic data in the verification set, three primary learners are respectively used for prediction to obtain the demand characteristic data x in the verification set d The prediction result is denoted as h d ＝(h 1 (x d ),h 2 (x d ),h 3 (x d )) T D=1, 2, …, k, where k is the number of samples contained in the validation set, h 1 (x d )、h 2 (x d )、h 3 (x d ) Each representing a primary learner for demand characteristic data x d Is a predicted result of (a).
9. The method for predicting demand for response to significant outside variations as recited in claim 1 or 8, wherein predicting the amount of demand for the product using the linear hybrid total model comprises:
and taking the future demand characteristic data in the n+1 stage as input, respectively predicting by using three primary learners to obtain three primary prediction results, and taking the three primary prediction results as input of a linear hybrid total model, wherein the linear hybrid total model predicts the product demand quantity by weighting and fusing the three primary prediction results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341015.4A CN113240359B (en) | 2021-03-30 | 2021-03-30 | Demand prediction method for coping with external major changes |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110341015.4A CN113240359B (en) | 2021-03-30 | 2021-03-30 | Demand prediction method for coping with external major changes |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113240359A CN113240359A (en) | 2021-08-10 |
| CN113240359B true CN113240359B (en) | 2024-02-23 |
Family
ID=77130742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110341015.4A Active CN113240359B (en) | 2021-03-30 | 2021-03-30 | Demand prediction method for coping with external major changes |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113240359B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113887026A (en) * | 2021-09-17 | 2022-01-04 | 中国石油大学(华东) | Oil pipeline hydraulic prediction method based on ensemble learning |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102385724A (en) * | 2010-08-27 | 2012-03-21 | 上海财经大学 | Spare part assembling demand forecasting information processing method applied to inventory management |
| CN109685583A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of supply chain needing forecasting method based on big data |
| CN109934408A (en) * | 2019-03-18 | 2019-06-25 | 常伟 | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning |
| CN110751327A (en) * | 2019-10-17 | 2020-02-04 | 武汉大学 | Long-term load combination prediction method based on multiple linear regression and gray Verhulst model |
| CN111191823A (en) * | 2019-12-20 | 2020-05-22 | 西北工业大学 | Production logistics prediction method based on deep learning |
| CN111401808A (en) * | 2020-03-12 | 2020-07-10 | 重庆文理学院 | Material agreement inventory demand prediction method based on hybrid model |
| CN112288191A (en) * | 2020-11-19 | 2021-01-29 | 国家海洋信息中心 | Ocean buoy service life prediction method based on multi-class machine learning method |
| CN112488392A (en) * | 2020-12-01 | 2021-03-12 | 重庆邮电大学 | Intelligent water affair daily water consumption prediction method based on machine learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130246017A1 (en) * | 2012-03-14 | 2013-09-19 | Microsoft Corporation | Computing parameters of a predictive model |
| CN107133699A (en) * | 2017-05-12 | 2017-09-05 | 百度国际科技（深圳）有限公司 | Box office receipts trend forecasting method and device, equipment, storage medium |
-
2021
- 2021-03-30 CN CN202110341015.4A patent/CN113240359B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102385724A (en) * | 2010-08-27 | 2012-03-21 | 上海财经大学 | Spare part assembling demand forecasting information processing method applied to inventory management |
| CN109685583A (en) * | 2019-01-10 | 2019-04-26 | 博拉网络股份有限公司 | A kind of supply chain needing forecasting method based on big data |
| CN109934408A (en) * | 2019-03-18 | 2019-06-25 | 常伟 | A kind of application analysis method carrying out automobile batteries RUL prediction based on big data machine learning |
| CN110751327A (en) * | 2019-10-17 | 2020-02-04 | 武汉大学 | Long-term load combination prediction method based on multiple linear regression and gray Verhulst model |
| CN111191823A (en) * | 2019-12-20 | 2020-05-22 | 西北工业大学 | Production logistics prediction method based on deep learning |
| CN111401808A (en) * | 2020-03-12 | 2020-07-10 | 重庆文理学院 | Material agreement inventory demand prediction method based on hybrid model |
| CN112288191A (en) * | 2020-11-19 | 2021-01-29 | 国家海洋信息中心 | Ocean buoy service life prediction method based on multi-class machine learning method |
| CN112488392A (en) * | 2020-12-01 | 2021-03-12 | 重庆邮电大学 | Intelligent water affair daily water consumption prediction method based on machine learning |
Non-Patent Citations (2)
| Title |
|---|
| 基于专利技术交易数据的京津冀区域 热点需求技术识别研究;程力培 等;《农业图书情报学刊》;第30卷(第7期);第5-9页 * |
| 改进的时间相关序列股票价格混合预测模型研究;王晓红;王梦瑶;郝婷;;科技促进发展(第06期);第82-88页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113240359A (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Predicting construction cost and schedule success using artificial neural networks ensemble and support vector machines classification models | |
| Patriarca et al. | Inventory model for a multi-echelon system with unidirectional lateral transshipment | |
| Lughofer et al. | On employing fuzzy modeling algorithms for the valuation of residential premises | |
| CN111080338B (en) | User data processing method and device, electronic equipment and storage medium | |
| Rezaei et al. | A neural network approach for retailer risk assessment in the aftermarket industry | |
| Boateng et al. | A dynamic framework for managing the complexities of risks in megaprojects | |
| Vahabi et al. | A sales forecasting model in automotive industry using Adaptive Neuro-Fuzzy Inference System (ANFIS) and Genetic Algorithm (GA) | |
| Relich | A computational intelligence approach to predicting new product success | |
| CN113656691A (en) | Data prediction method, device and storage medium | |
| Guo et al. | Reinforcement learning enabled dynamic bidding strategy for instant delivery trading | |
| CN113240359B (en) | Demand prediction method for coping with external major changes | |
| KR102554135B1 (en) | Method for providing product discount coupons based on store demand forecasting and system thereof | |
| Zhang | A deep learning model for ERP enterprise financial management system | |
| Camelia et al. | A Computational Grey Based Model for Companies Risk Forecasting. | |
| Li et al. | Multiple linear regression with Kalman filter for predicting end prices of online auctions | |
| US20220138887A1 (en) | Method and system for constructing virtual environment for ride-hailing platforms | |
| US11288691B2 (en) | Systems and methods for price markdown optimization | |
| Kauko | Towards the 4th generation-an essay on innovations in residential property value modelling expertise | |
| KR20230109070A (en) | Method for determination of credit evaluation score card based on e-commerce data and apparatus for performing the method | |
| Jackson et al. | Automl approach to classification of candidate solutions for simulation models of logistic systems | |
| Rymarczyk et al. | Profiling and segmenting clients with the use of machine learning algorithms | |
| Shah et al. | Predictive Analytic Modeling: A Walkthrough | |
| Chashmi et al. | Predicting customer turnover using recursive neural networks | |
| KR101927317B1 (en) | Method and Server for Estimating Debt Management Capability | |
| Aguilar et al. | Cost management in the digital era |
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 |