CN109785003A - A kind of Pharmaceutical retail industry medicine sales forecasting system and method - Google Patents
A kind of Pharmaceutical retail industry medicine sales forecasting system and method Download PDFInfo
- Publication number
- CN109785003A CN109785003A CN201910044839.8A CN201910044839A CN109785003A CN 109785003 A CN109785003 A CN 109785003A CN 201910044839 A CN201910044839 A CN 201910044839A CN 109785003 A CN109785003 A CN 109785003A
- Authority
- CN
- China
- Prior art keywords
- data
- drug
- sales
- prediction
- sales volume
- 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
- 239000003814 drug Substances 0.000 title claims abstract description 214
- 238000000034 method Methods 0.000 title claims abstract description 85
- 229940079593 drug Drugs 0.000 claims abstract description 169
- 238000009499 grossing Methods 0.000 claims abstract description 56
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 238000007405 data analysis Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 30
- 238000010606 normalization Methods 0.000 claims description 18
- 238000004458 analytical method Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 10
- 229940126532 prescription medicine Drugs 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 238000005096 rolling process Methods 0.000 claims description 7
- 238000013480 data collection Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000001186 cumulative effect Effects 0.000 claims description 4
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 3
- 238000013450 outlier detection Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 210000004218 nerve net Anatomy 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 4
- 210000005036 nerve Anatomy 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012731 temporal analysis Methods 0.000 description 2
- 238000000700 time series analysis Methods 0.000 description 2
- 241000254158 Lampyridae Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention discloses a kind of Pharmaceutical retail industry medicine sales forecasting systems, belong to the technical field of big data and artificial intelligence, including data acquisition and pretreatment unit, historical data analysis unit and predicting unit, additionally provide a kind of Pharmaceutical retail industry medicine sales prediction technique, comprising: data acquisition and pretreatment；Historical data analysis and data normalized；Data prediction: establishing Three-exponential Smoothing prediction model, grey forecasting model, LSTM prediction model respectively, constitutes BP neural network combination forecasting by Three-exponential Smoothing prediction model, grey forecasting model, the combination of LSTM prediction model；It is final to obtain the combined prediction result based on BP neural network, prediction model is fast and accurately established according to historic sales data to reach, the sales data of each pharmacy drug of following a period of time is predicted, and then provide data for storehouse drug deposit and support, to meet the purpose of the drug equilibrium of supply and demand.
Description
Technical field
The invention belongs to the technical fields of big data and artificial intelligence, in particular to a kind of Pharmaceutical retail industry medicine
Product sales forecasting system and method.
Background technique
To guarantee that the normal operation of drugs supply system can reduce the operation costs such as inventory to the greatest extent, also can guarantee between continuing not
Disconnected drugs supply.Realize above-mentioned two target, it is necessary to which the demand to every kind of drug carries out Accurate Prediction.If prediction is much
Higher than actual demand amount, then it certainly will lead to very high inventory level, thus the operation costs such as stock buildup；, whereas if prediction
Quantity is too low, then may cause frequent shortage of goods.
Method for Sales Forecast is using Method for Sales Forecast model according to previous sales volume situation to obtain to the pre- of the following sales volume situation
It surveys.Currently, Method for Sales Forecast mostly uses greatly the method based on time series analysis to be predicted, time series analysis method was used
The historical data gone further speculates following development trend by statistical analysis, its foundation is that all things are all development
Variation, the development and change of things have continuity in time.Due to the relevant complexity of field of medicaments, lack for drug
The analysis and prediction technique of sales volume.
In Patent No.: CN104951843B, denomination of invention are as follows: according to the area of new region in Method for Sales Forecast system and method
The degree of correlation matching degree of the area information in domain information and other regions predicts this in the sales volume in other regions according to certain product
Sales volume of the product in a new region.But it is not described certain product in the Method for Sales Forecast method in other regions, it needs to know
One, road product could predict the product in the sales volume of a new region in the Method for Sales Forecast value of known region.
In Patent No.: CN106779797A, denomination of invention are as follows: be based on inverse proportion polynomial function firefly optimization algorithm
Support vector machines drug prediction technique in, do not account for the relationship in sequence between context, this point is to time series data
For, it is particularly important.
In Patent No.: 108985483 A of CN, denomination of invention are as follows: initially set up in a kind of method of drug Method for Sales Forecast
Essential information relationship calculates the characteristic index of patient, and carries out machine learning and classify automatically to patient.Then it uses
ARIMA model predicts every a kind of patient, the quantity of every a kind of patient in the coming year is obtained, according to every a kind of patient
Prediction, obtain the predicted value that the drug or medical instrument are formed based on disease.The model only has patient assessment's record to hospital
The case where be relatively applicable in, not applicable for the drug Method for Sales Forecast of pharmacy, pharmacy does not record the information and medical note of patient
Record.
Therefore, a kind of drug Method for Sales Forecast method that precision of prediction is high is needed.
Summary of the invention
In view of this, in order to solve the above problems existing in the present technology, the purpose of the present invention is to provide a kind of drugs
Retail trade medicine sales forecasting system and method fast and accurately establish prediction model according to historic sales data to reach,
The sales data of each pharmacy drug of following a period of time is predicted, and then provides data for storehouse drug deposit and supports,
To meet the purpose of the drug equilibrium of supply and demand.
The technical scheme adopted by the invention is as follows: a kind of Pharmaceutical retail industry medicine sales forecasting system, including data are adopted
Collection and pretreatment unit, historical data analysis unit, predicting unit, the data acquisition and pretreatment unit and the history number
It communicates to connect according between analytical unit, is communicated to connect between historical data analysis unit and the predicting unit；
For acquiring drug historic sales data, drug historic sales data includes for the data acquisition and pretreatment unit
Drug history sales volume data, and drug historic sales data is pre-processed, selection index system is in the basic data of trend analysis
Collection；
The historical data analysis unit is used to carry out drug historic sales data outlier detection and processing, normalization
Processing, to obtain normalization data；
The predicting unit is used for normalization data according to drug history sales volume data and basic data collection, to the following medicine
Product sales volume is predicted.
The present invention also provides a kind of Pharmaceutical retail industry medicine sales prediction techniques, comprising the following steps:
(1) data acquisition and pretreatment: the medicine sales data of past M are collected, and after being pre-processed, are
Trend analysis selects time cycle m；
(2) historical data analysis:
(2.1) outlier identification and processing: exceptional value is identified using linear regression and deleted residual method, and using adjacent
The method of average value replaces the exceptional value in two periods；
(2.2) data normalization is handled: being made normalized to all data in medicine sales data, and is made to upset place
Reason；
(3) data are predicted:
(3.1) Three-exponential Smoothing prediction model, grey forecasting model, LSTM prediction model are established respectively, and are counted respectively
It calculates and obtains Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result；
(3.2) BP nerve is constituted by Three-exponential Smoothing prediction model, grey forecasting model, the combination of LSTM prediction model
Combination of network prediction model；
(3.3) BP neural network combination forecasting is pre- with Three-exponential Smoothing prediction result, gray prediction result, LSTM
Result is surveyed as input variable, and model training is carried out to BP neural network combination forecasting, it is final to obtain based on BP nerve
The combined prediction result of network.
Further, in the step (1), the medicine sales data include drug sales volume data, the drug sales volume number
It whether is festivals or holidays, the drug sales volume on the day of according to corresponding month, corresponding all several, the drug sales volume data of the drug sales volume data
The drug of data whether be prescription medicine, the drug sales volume data corresponding same day temperature warp corresponding with the drug sales volume data
Latitude.
Further, in the step (2.2), by medicine sales data by formula (1) be nondestructively mapped to [0,1] it
Between, formula (1) is as follows:
Further, drug sales volume data, the drug sales volume data corresponding month, the medicine in the medicine sales data
Product sales volume data corresponding week is several, on the day of the drug sales volume data whether be festivals or holidays, the drug sales volume data drug whether be
The temperature longitude and latitude corresponding with the drug sales volume data on prescription medicine, the drug sales volume data corresponding same day is used as because of prime number
According to using each factor data as feature, the normalized is that each feature is normalized respectively.
Further, the input of the Three-exponential Smoothing prediction model are as follows: the drug sales volume data of certain drug of past N days
{x1,x2,...,xN}；Output are as follows: N+1, N+2 ..., N+j days drug Method for Sales Forecast values, calculating process is as follows:
(a1) S ' is determined1Value:
S′1Indicate the initial value of the single exponential smoothing value of the 1st day drug sales volume in training number of days, S '1Value take the 1st day
Drug sales volume x1, formula is as follows:
S′1=x1 (2)
(a2) α=0.1 is enabled；
(a3) single exponential smoothing of each observing time point below is calculated according to formula (3-1), (3-2), (3-3)
Value S 'tWith double smoothing value S "tWith Three-exponential Smoothing value S " 't；
S′t=α xt+(1-α)S′t-1 (3-1)
S″t=α S 't+(1-α)S″t-1 (3-2)
S″′t=α S "t+(1-α)S″′t (3-3)
(a4) Three-exponential Smoothing predicted value is calculated:
Xt=At+Bt+Ct, t=2 ..., N (4)
Wherein, At=3S 't-3S″t+S″′t,
(a5) prediction of the Three-exponential Smoothing under α error RMSE is calculated according to formula (5):
Wherein, N indicates the observation number of days of drug sales volume data；
(a6) α=α+0.1 returns to (a3) step, computes repeatedly, until α=0.9；
(a7) it asks so that predict the minimum corresponding α of error, is denoted as αbest；
(a8) according to formula (2), (3-1), (3-2), (3-3) and (4) respectively obtain the t days single exponential smoothing value,
Double smoothing value, Three-exponential Smoothing value and exponential forecasting value X three timest；
(a9) one day initial data to be predicted is calculated；
By one day initial data to be predicted be set as the previous day, a few days ago, the weighting of the last week, the previous moon, formula
It is as follows:
xN+1=w1xN+w2xN-1+w3xN-2+w4xN-T+1″ (6)
Wherein, w1+w2+w3+w4=1, T are according to the periodically selected of data, and T=m；
(a10) the N+1 days value of exponential forecasting three times X are calculated separately using formula (4)N+1；
(a11) the Three-exponential Smoothing predicted value of step (a9) to (a10) until all number of days to be predicted have been calculated is repeated.
Further, the input of the grey forecasting model are as follows: the drug sales volume data { x of certain drug of past N days1,
x2,...,xN}；Output are as follows: N+1, N+2 ..., N+j days drug Method for Sales Forecast values, the calculating process of grey forecasting model is such as
Under:
(b1) by drug sales volume data { x1,x2,...,xNIt is denoted as: X(0)={ x(0)(1),x(0)(2),...,x(0)(N) },
Calculate one-accumulate sequence X(1)={ x(1)(1), x(1)(2) ..., x(1)(N)}；
(b2) matrix B, y are established；
YN=[x(0)(2),x(0)(3),...,x(0)(N)]T (7-2)
(b3) finding the inverse matrix (BTB)-1；
(b4) basisSeek estimated valueWith
(b5) prediction model is obtained:
With prediction model digital simulation valueSubtract operation reduction after using again, i.e.,
(b6) accuracy test:
According to prediction model, accuracy test is carried out to each match value, if forecast ratings are good, according to step
(b5) prediction model obtained enters step (b7) and is predicted；Otherwise, (b8) is entered step；
(b7) it predicts:
It is obtained according to formula (7-4)Subtract operation after, obtains N+1 days drug Method for Sales Forecast
Value, it may be assumed that
(b8) the highest model of coincidence factor is sought:
Successively calculate secondary cumulative sequence X(2)={ x(2)(1), x(2)(2) ..., x(2)(N) }, add up sequence X three times(3)=
{x(3)(1), x(3)(2) ..., x(3)(N) } it, is calculated according to step (b2) to (b7)The precision and relative error of different models are obtained, curve is analyzed, selection is pre-
The highest model of precision is surveyed, subtracts operation after, obtains N+1 days drug Method for Sales Forecast values, it may be assumed that
Further, the specific method is as follows for the LSTM prediction model: being realized using Python code based on Keras's
LSTM multivariable drug Method for Sales Forecast；The data set that medicine sales data conversion is predicted at Multivariate Time Series are suitable for,
Using the method for rolling forecast, the training set of building LSTM network is rolled, inputs the LSTM network put up, LSTM network passes through
The feature of learning data updates network parameter and weight, and exports following one day drug sales volume data.
Further, the calculation method of the BP neural network combination forecasting is as follows:
(d1) training stage: the data of Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result are returned
Input after one change as BP neural network combination, and practical drug sales volume data normalization is the defeated of BP neural network combination
Out, the BP neural network training based on Keras is realized using Python code, obtain training pattern；
(d2) forecast period: drug sales volume data are predicted using the above trained training pattern, and enable input
Layer are as follows:
y1t: third index flatness prediction result；
y2t: gray prediction result；
y3t: LSTM prediction result；
Then output layer are as follows:
yt: the combined prediction result based on BP neural network.
The invention has the benefit that
1. Pharmaceutical retail industry medicine sales prediction technique disclosed in this invention, since it is to improve to refer to three times in foundation
On the basis of number smoothing prediction model, grey forecasting model, multivariable LSTM prediction model, three of the above Individual forecast mould is utilized
The combination of type constitutes BP neural network combination forecasting, and when preventing training over-fitting, Individual forecast method causes precision of prediction
Not high problem；It is sent by the prediction result of comprehensive Three-exponential Smoothing, the drug sales volume data of gray prediction, LSTM prediction
BP neural network obtains the optimal combinatorial forecasting of three kinds of models as a result, improving the prediction precision of drug sales volume.
2. in Pharmaceutical retail industry medicine sales prediction technique provided by the present invention, LSTM network be suitable for processing with
Very long critical event is spaced and postponed in predicted time sequence, is carried out drug Method for Sales Forecast using LSTM network, is fitted energy
Power and prediction effect all show great superiority.
3. fixed using rolling forecast method in Pharmaceutical retail industry medicine sales prediction technique provided by the present invention
History real data for the previous period is included in prediction and considered by the phase, to respond market for product actual demand, monthly one produces
New sales data is given birth to it is necessary to be adjusted to following prediction data according to the actual condition of sales in market；Not only improve
Predictablity rate can quickly discover turn of the market, start strain mechanism ahead of time, reserve the preposition of abundance for production, logistics
Time allows supply chain to support the medium- and long-term strategy planning for being also beneficial to enterprise more in place, makes more thorough decision and examine
Amount.
Detailed description of the invention
Fig. 1 is the system structure diagram of Pharmaceutical retail industry medicine sales forecasting system provided by the invention；
Fig. 2 is the flow diagram of Pharmaceutical retail industry medicine sales prediction technique provided by the invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.The present invention being usually described and illustrated herein in the accompanying drawings is implemented
The component of example can be arranged and be designed with a variety of different configurations.
Therefore, the detailed description of the embodiment of the present invention provided in the accompanying drawings is not intended to limit below claimed
The scope of the present invention, but be merely representative of selected embodiment of the invention.Based on the embodiments of the present invention, this field is common
Technical staff's every other embodiment obtained without creative efforts belongs to the model that the present invention protects
It encloses.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the present invention can phase
Mutually combination；And all model trainings are programmed on eight generation of Intel's Duo (6 cores, 6 threads) processor using Python
It completes.
As shown in Figure 1, the present invention provides a kind of Pharmaceutical retail industry medicine sales forecasting system, which includes
Data acquisition and pretreatment unit, historical data analysis unit, predicting unit, data acquisition and pretreatment unit with it is described
It communicates to connect between historical data analysis unit, is communicated to connect between historical data analysis unit and the predicting unit；
For acquiring drug historic sales data, drug historic sales data includes for the data acquisition and pretreatment unit
Drug history sales volume data, and drug historic sales data is pre-processed, selection index system is in the basic data of trend analysis
Collection；
The historical data analysis unit is used to carry out drug historic sales data outlier detection and processing, normalization
Processing, to obtain normalization data；
The predicting unit is used for normalization data according to drug history sales volume data and basic data collection, to the following medicine
Product sales volume is predicted, is predicted with the sales data of each pharmacy drug to following a period of time, and then is storehouse drug
Deposit provides data and supports, so that meeting the drug equilibrium of supply and demand as far as possible.
As shown in Fig. 2, the present invention provides a kind of Pharmaceutical retail industry medicine sales prediction technique, the prediction technique is specific
The following steps are included:
(1) data acquisition and pretreatment: the medicine sales data of past M are collected, and after being pre-processed, root
It include the quantity of historical data point required for being obtained in trend analysis according to preset time cycle m, selection index system is in trend analysis
Basic data collection, it is preferred that in embodiments of the present invention, we select time cycle be 7 days, i.e. m=7；
In the step, the medicine sales data include drug sales volume data, the drug sales volume data corresponding month, are somebody's turn to do
Drug sales volume data corresponding week is several, on the day of the drug sales volume data whether be festivals or holidays, the drug sales volume data drug whether
For the temperature longitude and latitude corresponding with the drug sales volume data on the same day corresponding to prescription medicine, the drug sales volume data.
By above-mentioned drug sales volume data, the drug sales volume data corresponding month, the drug sales volume data corresponding week it is several,
On the day of drug sales volume data whether be festivals or holidays, the drug sales volume data drug whether be prescription medicine, the drug sales volume number
According to the temperature on corresponding same day longitude and latitude corresponding with the drug sales volume data as the factor data for influencing drug sales volume.
(2) historical data analysis:
(2.1) outlier identification and processing: exceptional value is identified using linear regression and deleted residual method, and using adjacent
The method of average value replaces the exceptional value in two periods；
(2.2) data normalization is handled: being made normalized to all data in medicine sales data, and is made to upset place
Reason, specifically, medicine sales data are nondestructively mapped between [0,1] by formula (1), formula (1) is as follows:
In step (1), each factor data is as each feature in prediction model, due to different influence factors
Data magnitudes is different, and according to the same method for normalizing, then data smaller for data magnitude, data become after normalization
Very little is obtained, and prediction result is influenced less, therefore, normalization processing method is returned respectively to each feature in the step
One changes, it may be assumed that by drug sales volume data, the drug sales volume data corresponding month, the corresponding all several, drugs of the drug sales volume data
It whether is whether festivals or holidays, the drug of the drug sales volume data are right by prescription medicine, the drug sales volume data on the day of sales volume data
Should day temperature longitude and latitude corresponding with the drug sales volume data as factor data, using each factor data as spy
Sign, the normalized is that each feature is normalized respectively.
(3) data are predicted:
(3.1) Three-exponential Smoothing prediction model, grey forecasting model, LSTM prediction model are established respectively, and are counted respectively
It calculates and obtains Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result；
(a) it is calculated by Three-exponential Smoothing prediction model and obtains Three-exponential Smoothing prediction result:
The input of Three-exponential Smoothing prediction model are as follows: the drug sales volume data { x of certain drug of past N days1,x2,...,
xN}；Output are as follows: N+1, N+2 ..., N+j days drug Method for Sales Forecast values, wherein j represents the number of days that predict several days of future,
It can select according to actual needs, calculating process is as follows:
(a1) S ' is determined1Value:
S′1Indicate the initial value of the single exponential smoothing value of the 1st day drug sales volume in training number of days, S '1Value take the 1st day
Drug sales volume x1, formula is as follows:
S′1=x1 (2)
(a2) α=0.1 is enabled；
(a3) single exponential smoothing of each observing time point below is calculated according to formula (3-1), (3-2), (3-3)
Value S 'tWith double smoothing value S "tWith Three-exponential Smoothing value S " 't；
S′t=α xt+(1-α)S′t-1 (3-1)
S″t=α S 't+(1-α)S″t-1 (3-2)
S″′t=α S "t+(1-α)S″′t (3-3)
(a4) Three-exponential Smoothing predicted value is calculated:
Xt=At+Bt+Ct, t=2 ..., N (4)
Wherein, At=3S 't-3S″t+S″′t,
(a5) prediction of the Three-exponential Smoothing under α error RMSE is calculated according to formula (5):
Wherein, N indicates the observation number of days of drug sales volume data；
(a6) α=α+0.1 returns to (a3) step, computes repeatedly, until α=0.9；
(a7) it asks so that predict the minimum corresponding α of error, is denoted as αbest；
(a8) according to formula (2), (3-1), (3-2), (3-3) and (4) respectively obtain the t days single exponential smoothing value,
Double smoothing value, Three-exponential Smoothing value and exponential forecasting value X three timest；
(a9) one day initial data to be predicted is calculated；
By one day initial data to be predicted be set as the previous day, a few days ago, the weighting of the last week, the previous moon, formula
It is as follows:
xN+1=w1xN+w2xN-1+w3xN-2+w4xN-T+1″ (6)
Wherein, w1+w2+w3+w4=1, T are according to the periodically selected of data, and T=m, it is preferred that m=7；
(a10) the N+1 days value of exponential forecasting three times X are calculated separately using formula (4)N+1；
(a11) the Three-exponential Smoothing predicted value of step (a9) to (a10) until all number of days to be predicted have been calculated is repeated.
(b) it is calculated according to grey forecasting model and obtains gray prediction result:
Grey forecasting model uses general grey forecasting model, the input of grey forecasting model are as follows: certain drug of past N days
Drug sales volume data { x1,x2,...,xN}；Output are as follows: N+1, N+2 ..., N+j days drug Method for Sales Forecast values, wherein j generation
Table will predict following several days number of days, can select according to actual needs, the calculating process of grey forecasting model is as follows:
(b1) by drug sales volume data { x1,x2,...,xNIt is denoted as: X(0)={ x(0)(1),x(0)(2),...,x(0)(N) },
Calculate one-accumulate sequence X(1)={ x(1)(1), x(1)(2) ..., x(1)(N)}；
(b2) matrix B, y are established；
YN=[x(0)(2),x(0)(3),...,x(0)(N)]T (7-2)
(b3) finding the inverse matrix (BTB)-1；
(b4) basisSeek estimated valueWith
(b5) prediction model is obtained:
With prediction model digital simulation valueSubtract operation reduction after using again, i.e.,
(b6) accuracy test:
According to prediction model, accuracy test is carried out to each match value and is obtained if forecast ratings are good according to step (b5)
Prediction model enter step (b7) and predicted；Otherwise, (b8) is entered step；
Accuracy test model in the step is inspection method of accuracy general in grey forecasting model, is specifically included following
Step:
(b6.1) it residual test: calculates separately
Residual error:
Opposite residual error:
(b6.2) it posterior difference examination: calculates separately
x(0)Mean value:
x(0)Variance:
The mean value of residual error:
The variance of residual error:
Posteriority difference ratio:Small error possibility:
(b6.3) precision of prediction grade
It indicates that precision of prediction is good when P>0.95, C<0.35, when 0.8<P≤0.95,0.35≤C<0.45, indicates that prediction is closed
Lattice when 0.7 < P≤0.8,0.45≤C < 0.5, indicate prediction reluctantly, otherwise indicate that prediction is unqualified.
(b7) it predicts:
It is obtained according to formula (7-4)Subtract operation after, obtains N+1 days drug Method for Sales Forecast
Value, it may be assumed that
(b8) the highest model of coincidence factor is sought:
Successively calculate secondary cumulative sequence X(2)={ x(2)(1), x(2)(2) ..., x(2)(N) }, add up sequence X three times(3)=
{x(3)(1), x(3)(2) ..., x(3)(N) } it, is calculated according to step (b2) to (b7)The precision and relative error of different models are obtained, curve is analyzed, selection is pre-
The highest model of precision is surveyed, subtracts operation after, obtains N+1 days drug Method for Sales Forecast values, it may be assumed that
The circular of the step is as follows: rolling forecast N+2, N+3...N+j days drug Method for Sales Forecast values
If it is corresponding to be obtained the highest model of precision of prediction by the corresponding precision of prediction highest of one-accumulate for predictionOriginal series are added, the N+2 days drug Method for Sales Forecast values obtained according to step (1) to (7) prediction are as grey
The drug Method for Sales Forecast value of prediction modelThen again willOriginal series are added, successively obtain
If the secondary corresponding precision of prediction highest that adds up, calculates secondary cumulative sequence X first(2)={ x(2)(1), x(2)
(2) ..., x(2)(N), x(2)(N+1) } it, is then calculated according to step (2) to (7)In turn
N+2 days drug Method for Sales Forecast values are calculated, it may be assumed thatThen again willOriginal series are added, successively
It obtains
If adding up corresponding precision of prediction highest three times, the sequence X that adds up three times is calculated first(3)={ x(3)(1), x(3)
(2) ..., x(3)(N), x(3)(N+1) } it, is then calculated according to step (2) to (7)In turn
N+2 days drug Method for Sales Forecast values are calculated, it may be assumed thatThen again willOriginal series are added, according to
It is secondary to obtain
(c) it is calculated according to LSTM prediction model and obtains LSTM prediction result:
The specific method is as follows for LSTM prediction model: realizing the LSTM multivariable medicine based on Keras using Python code
Product Method for Sales Forecast；Basic data collection is converted into be suitable for the data set of Multivariate Time Series prediction, uses rolling forecast
Method rolls the training set of building LSTM network, inputs the LSTM network put up, the feature that LSTM network passes through learning data
Network parameter and weight are updated, and exports following one day drug sales volume data.Wherein, LSTM network is a kind of improved time
Recognition with Recurrent Neural Network, LSTM can be single due in neural network including time memory with learning time sequence shot and long term Dependency Specification
Member, thus suitable for handle and predicted time sequence interval and delay event.
In the prior art, BP neural network, support vector machines are all to find drug sales volume shadow using the method for machine learning
The Nonlinear Mapping relationship between factor and drug sales volume is rung, and has ignored sequence data between continuous drug sales volume sample
Correlation.In fact, as typical time series, drug sales volume data not only have non-linear, and have correlation,
I.e. for given area, the variation of drug sales volume data is a continuous process, current drug sales volume and the previous day
It is not independent from each other between drug sales volume, there is very strong correlations between the two.So daily drug sales volume becomes
Change depends not only upon same day input feature vector, and related with past input feature vector.Therefore, conventional method is only to single sample
Non-linear relation is established in input feature vector and output, the strong correlation being lost between continuous sequence sample, and precision of prediction is limited.
The method of LSTM prediction model is specific as follows:
(1.1) the selected principal element for influencing target of grey relational grade analysis: it will affect the p kind factor of drug sales volume factor
Sequence carries out grey relational grade analysis with drug sales volume sequence respectively, selects the q kind factor that the degree of association is greater than given value.
In objective world, the relationship between many factors is grey, and it is close to be hard to tell which factor relation, which because
Element is far, is just difficult to find that principal contradiction, finds main feature, prevailing relationship.Correlation analysis is in analysis system
The method of each correlate degree, needs first calculate correlation coefficient, then calculating correlation again.
If referential is classified as: X0={ X0(1),X0(2),...,X0(n)}；
Compared sequence are as follows: Xi={ Xi(1),Xi(2),...,Xi(n) }, i=1,2 ..., k；
1.1.1 it) initializes
Different to unit, the different sequence of initial value should be initialized first before calculate correlation coefficient, i.e., by the sequence
All data are respectively divided by the first data.
1.1.2 absolute difference sequence) is sought
Δi(t)=| X0(t)-Xi(t) |, t=1,2 ..., n (8)
1.1.3) seek incidence coefficient
It indicates to refer to series of X0With i-th of series of X by compared withiAbsolute difference at t point, then incidence coefficient is defined as:
In formula, ρ is resolution ratio, between zero and one, generally takes 0.5.
1.1.4) the degree of association
Compared sequence XiWith reference series of X0The degree of association both be defined as the average value of each point incidence coefficient, that is, be associated with
Degree:
(1.2) LSTM network query function
According to degree of association selected characteristic value: the drug sales volume data, day to be predicted in one 7 days time cycle of past are corresponding
Month, the corresponding week day several, to be predicted in day to be predicted whether be weekend, whether the drug is prescription medicine, drug sales volume data pair
The temperature and the corresponding longitude and latitude of drug sales volume data on the same day answered, totally 14 characteristic values select the degree of association big according to the degree of association
In the p kind feature of given value, in embodiments of the present invention, p=12.
It chooses 1000 days history characteristics in the past and is respectively as follows: 7 days corresponding past drug history pin as input
Sell data, corresponding month, corresponding week it is several, whether be weekend, whether the drug is prescription medicine, corresponding temperature, longitude and latitude etc.
The p kind feature that the degree of association is greater than given value in information and is trained and predicts as LSTM network inputs.
Network maximum frequency of training 1000；
E-learning rate 0.01；
The implicit number of plies: 2；
Hidden layer neuron number: 30*30；
Input layer and hidden layer add dropout, and dropout=0.1；
Training stage (1000, sample):
Input: 1000*p, output: 1000*1；
Forecast period:
Input: 1*p, output: 1*1.
(1.3) rolling forecast
Using the mode of rolling forecast, the drug sales volume data for predicting one day to be predicted in the form of per next, then
Input value is added in the prediction drug sales volume data and relevant other information, predicts next day drug sales volume data, according to
It is secondary to analogize.
(1.4) model incremental updates
In order to guarantee the timeliness of model, need periodically to be updated LSTM prediction model, time interval is usually 3-6
A month (or even 1-2 month), if removing re -training LSTM prediction model using total data every time, time overhead is very big, because
This, can be trained LSTM prediction model by the way of incremental update LSTM prediction model, be assembled for training using master data
After having practiced LSTM prediction model, LSTM prediction model is first serialized, then import LSTM prediction model again and carries out increment instruction
Practice.
(3.2) BP nerve is constituted by Three-exponential Smoothing prediction model, grey forecasting model, the combination of LSTM prediction model
Combination of network prediction model；Using the prediction result of above-mentioned Individual forecast model as the input of network, with practical drug sales volume number
According to export, combined weight can be obtained according to the smallest criterion of mean square error by 3 layers of BP e-learning and training.
(3.3) BP neural network combination forecasting is pre- with Three-exponential Smoothing prediction result, gray prediction result, LSTM
Result is surveyed as input variable, and model training is carried out to BP neural network combination forecasting, it is final to obtain based on BP nerve
The combined prediction result of network.
The calculation method of the BP neural network combination forecasting is following (including training stage and forecast period):
(d1) training stage: the data of Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result are returned
Input after one change as BP neural network combination, and practical drug sales volume data normalization is the defeated of BP neural network combination
Out, the BP neural network training based on Keras is realized using Python code, obtain training pattern；
(d2) forecast period: drug sales volume data are predicted using the above trained training pattern, and enable input
Layer are as follows:
y1t: third index flatness prediction result；
y2t: gray prediction result；
y3t: LSTM prediction result；
Then output layer are as follows:
yt: the combined prediction result based on BP neural network；
If the precision of prediction for the grey forecasting model that step (d2) obtains is all unqualified, second variable is deleted, only
Utilize Three-exponential Smoothing prediction result y1t, LSTM prediction result y3tAs input variable, training BP neural network combination.
Establishing the basis for improving Three-exponential Smoothing prediction model, grey forecasting model, multivariable LSTM prediction model
On, BP neural network combination forecasting is constituted using the combination of three of the above Individual forecast model.It is quasi- in order to prevent to train
When conjunction, Individual forecast method causes the precision of prediction not high.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are various under the inspiration of the present invention
The product of form, however, make any variation in its shape or structure, it is all to fall into the claims in the present invention confining spectrum
Technical solution, be within the scope of the present invention.
Claims (9)
1. a kind of Pharmaceutical retail industry medicine sales forecasting system, which is characterized in that including data acquisition and pretreatment unit, go through
History data analysis unit, predicting unit are led between the data acquisition and pretreatment unit and the historical data analysis unit
Letter connection, communicates to connect between historical data analysis unit and the predicting unit；
For acquiring drug historic sales data, drug historic sales data includes drug for the data acquisition and pretreatment unit
History sales volume data, and drug historic sales data is pre-processed, selection index system is in the basic data collection of trend analysis；
The historical data analysis unit is used to carry out at outlier detection and processing, normalization drug historic sales data
Reason, to obtain normalization data；
The predicting unit is used for normalization data according to drug history sales volume data and basic data collection, to the following drug pin
Amount is predicted.
2. a kind of Pharmaceutical retail industry medicine sales prediction technique, which comprises the following steps:
(1) data acquisition and pretreatment: the medicine sales data of past M are collected, and after being pre-processed, for trend
Analysis selection time cycle m；
(2) historical data analysis:
(2.1) outlier identification and processing: exceptional value is identified using linear regression and deleted residual method, and using two neighboring
The method of average value replaces the exceptional value in period；
(2.2) data normalization is handled: being made normalized to all data in medicine sales data, and is made to upset processing；
(3) data are predicted:
(3.1) Three-exponential Smoothing prediction model, grey forecasting model, LSTM prediction model are established respectively, and are calculated separately and obtained
Take Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result；
(3.2) BP neural network is constituted by Three-exponential Smoothing prediction model, grey forecasting model, the combination of LSTM prediction model
Combination forecasting；
(3.3) BP neural network combination forecasting is with Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction knot
Fruit carries out model training as input variable, and to BP neural network combination forecasting, and final obtain is based on BP neural network
Combined prediction result.
3. Pharmaceutical retail industry medicine sales prediction technique according to claim 2, which is characterized in that the step (1)
In, the medicine sales data include drug sales volume data, the drug sales volume data corresponding month, the drug sales volume data pair
Answer week is several, on the day of drug sales volume data whether be festivals or holidays, the drug sales volume data drug whether be prescription medicine, the drug
The temperature longitude and latitude corresponding with the drug sales volume data on the sales volume data corresponding same day.
4. Pharmaceutical retail industry medicine sales prediction technique according to claim 2, which is characterized in that the step
(2.2) in, medicine sales data are nondestructively mapped between [0,1] by formula (1), formula (1) is as follows:
5. Pharmaceutical retail industry medicine sales prediction technique according to claim 4, which is characterized in that the medicine sales
Drug sales volume data, the drug sales volume data corresponding month, corresponding all several, the drug sales volumes of the drug sales volume data in data
On the day of data whether be festivals or holidays, the drug sales volume data drug whether be prescription medicine, work as corresponding to the drug sales volume data
It temperature longitude and latitude corresponding with the drug sales volume data is used as factor data, and enabling each factor data is respectively a feature,
The normalized is that each feature is normalized respectively.
6. Pharmaceutical retail industry medicine sales prediction technique according to claim 5, which is characterized in that the index three times
The input of smoothing prediction model are as follows: the drug sales volume data { x of certain drug of past N days1,x2,...,xN}；Output are as follows: N+1, N+
2 ..., N+j days drug Method for Sales Forecast values, calculating process are as follows:
(a1) S ' is determined1Value:
S′1Indicate the initial value of the single exponential smoothing value of the 1st day drug sales volume in training number of days, S '1Value take the 1st day medicine
Product sales volume x1, formula is as follows:
S′1=x1 (2)
(a2) α=0.1 is enabled；
(a3) the single exponential smoothing value S ' of each observing time point below is calculated according to formula (3-1), (3-2), (3-3)t
With double smoothing value S "tWith Three-exponential Smoothing value S " 't；
S′t=α xt+(1-α)S′t-1 (3-1)
S″t=α S 't+(1-α)S″t-1 (3-2)
S″′t=α S "t+(1-α)S″′t (3-3)
(a4) Three-exponential Smoothing predicted value is calculated:
Xt=At+Bt+Ct, t=2 ..., N (4)
Wherein, At=3S 't-3S″t+S″′t,
(a5) prediction of the Three-exponential Smoothing under α error RMSE is calculated according to formula (5):
Wherein, N indicates the observation number of days of drug sales volume data；
(a6) α=α+0.1 returns to (a3) step, computes repeatedly, until α=0.9；
(a7) it asks so that predict the minimum corresponding α of error, is denoted as αbest；
(a8) the t days single exponential smoothing values, secondary are respectively obtained according to formula (2), (3-1), (3-2), (3-3) and (4)
Exponential smoothing value, Three-exponential Smoothing value and exponential forecasting value X three timest；
(a9) one day initial data to be predicted is calculated；
By one day initial data to be predicted be set as the previous day, a few days ago, the weighting of the last week, the previous moon, formula is as follows:
xN+1=w1xN+w2xN-1+w3xN-2+w4xN-T+1″ (6)
Wherein, w1+w2+w3+w4=1, T is selected according to the periodicity of data, and T=m；
(a10) the N+1 days value of exponential forecasting three times X are calculated separately using formula (4)N+1；
(a11) the Three-exponential Smoothing predicted value of step (a9) to (a10) until all number of days to be predicted have been calculated is repeated.
7. Pharmaceutical retail industry medicine sales prediction technique according to claim 5, which is characterized in that the gray prediction
The input of model are as follows: the drug sales volume data { x of certain drug of past N days1,x2,...,xN}；Output are as follows: N+1, N+2 ..., N+j
It drug Method for Sales Forecast value, the calculating process of grey forecasting model are as follows:
(b1) by drug sales volume data { x1,x2,...,xNIt is denoted as: X(0)={ x(0)(1),x(0)(2),...,x(0)(N) } it, calculates
One-accumulate sequence X(1)={ x(1)(1), x(1)(2) ..., x(1)(N)}；
(b2) matrix B, y are established；
YN=[x(0)(2),x(0)(3),...,x(0)(N)]T (7-2)
(b3) finding the inverse matrix (BTB)-1；
(b4) basisSeek estimated valueWith
(b5) prediction model is obtained:
With prediction model digital simulation valueSubtract operation reduction after using again, i.e.,
(b6) accuracy test:
According to prediction model, accuracy test is carried out to each match value and is obtained according to step (b5) pre- if forecast ratings are good
Survey model enters step (b7) and is predicted；Otherwise, (b8) is entered step；
(b7) it predicts:
It is obtained according to formula (7-4)Subtract operation after, obtain N+1 days drug Method for Sales Forecast values,
That is:
(b8) the highest model of coincidence factor is sought:
Successively calculate secondary cumulative sequence X(2)={ x(2)(1), x(2)(2) ..., x(2)(N) }, add up sequence X three times(3)={ x(3)
(1), x(3)(2) ..., x(3)(N) } it, is calculated according to step (b2) to (b7)The precision and relative error of different models are obtained, curve is analyzed, selection is pre-
The highest model of precision is surveyed, subtracts operation after, obtains N+1 days drug Method for Sales Forecast values, it may be assumed that
8. Pharmaceutical retail industry medicine sales prediction technique according to claim 5, which is characterized in that the LSTM prediction
The specific method is as follows for model: realizing the LSTM multivariable drug Method for Sales Forecast based on Keras using Python code；By drug
Sales data is converted into being suitable for the data set of Multivariate Time Series prediction, using the method for rolling forecast, rolls building
The training set of LSTM network, inputs the LSTM network put up, and LSTM network updates network parameter by the feature of learning data
And weight, and export the drug sales volume data in one day future.
9. Pharmaceutical retail industry medicine sales prediction technique according to claim 2, which is characterized in that the BP nerve net
The calculation method of network combination forecasting is as follows:
(d1) training stage: by the data normalization of Three-exponential Smoothing prediction result, gray prediction result, LSTM prediction result
Input as BP neural network combination afterwards, and practical drug sales volume data normalization is the output of BP neural network combination, is made
The BP neural network training based on Keras is realized with Python code, obtains training pattern；
(d2) forecast period: drug sales volume data are predicted using the above trained training pattern, and enable input layer are as follows:
y1t: third index flatness prediction result；
y2t: gray prediction result；
y3t: LSTM prediction result；
Then output layer are as follows:
yt: the combined prediction result based on BP neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910044839.8A CN109785003A (en) | 2019-01-17 | 2019-01-17 | A kind of Pharmaceutical retail industry medicine sales forecasting system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910044839.8A CN109785003A (en) | 2019-01-17 | 2019-01-17 | A kind of Pharmaceutical retail industry medicine sales forecasting system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109785003A true CN109785003A (en) | 2019-05-21 |
Family
ID=66501622
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910044839.8A Pending CN109785003A (en) | 2019-01-17 | 2019-01-17 | A kind of Pharmaceutical retail industry medicine sales forecasting system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785003A (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110491120A (en) * | 2019-07-05 | 2019-11-22 | 广州车行易科技股份有限公司 | A kind of road traffic control method and its system based on big data |
| CN110599234A (en) * | 2019-08-13 | 2019-12-20 | 广东工业大学 | Product sales prediction method |
| CN110619418A (en) * | 2019-07-26 | 2019-12-27 | 重庆大学 | Multi-feature water quality prediction method based on mixed model combination algorithm |
| CN110717628A (en) * | 2019-10-09 | 2020-01-21 | 浪潮软件股份有限公司 | Goods source optimal distribution model construction method, optimal distribution model and optimal distribution method |
| CN111310977A (en) * | 2020-01-20 | 2020-06-19 | 重庆亚德科技股份有限公司 | Big data-based drug inventory prediction system and method |
| CN111428930A (en) * | 2020-03-24 | 2020-07-17 | 中电药明数据科技(成都)有限公司 | GBDT-based medicine patient using number prediction method and system |
| CN111445095A (en) * | 2020-05-18 | 2020-07-24 | 江苏电力信息技术有限公司 | Annual ex-warehouse amount prediction method based on exponential smoothing method |
| CN112001740A (en) * | 2020-06-19 | 2020-11-27 | 南京理工大学 | Combined prediction method based on adaptive neural network |
| CN112053233A (en) * | 2020-09-04 | 2020-12-08 | 天元大数据信用管理有限公司 | Dynamic small and medium enterprise credit scoring method and system based on GRA |
| CN112133415A (en) * | 2020-09-14 | 2020-12-25 | 微医云(杭州)控股有限公司 | Method, device, equipment and storage medium for replacing medicament |
| CN112148557A (en) * | 2020-09-15 | 2020-12-29 | 北京基调网络股份有限公司 | Method for predicting performance index in real time, computer equipment and storage medium |
| CN112820414A (en) * | 2021-01-29 | 2021-05-18 | 南威软件股份有限公司 | Early warning method for new crown epidemic situation based on improved cubic exponential smoothing model and LSTM model |
| CN113190535A (en) * | 2021-01-19 | 2021-07-30 | 云南卫士盾科技有限公司 | Gas concentration prediction method based on improved grey-long and short-term memory neural network combined model |
| CN113361972A (en) * | 2021-07-13 | 2021-09-07 | 壹药网科技(上海)股份有限公司 | Calculation system of dominant drugs |
| CN113724831A (en) * | 2021-08-31 | 2021-11-30 | 平安国际智慧城市科技股份有限公司 | Artificial intelligence-based medicine allocation method and device, terminal equipment and medium |
| CN113837794A (en) * | 2021-08-26 | 2021-12-24 | 润联软件系统(深圳)有限公司 | Chain retail store sales prediction method based on space-time graph convolutional network |
| CN114444946A (en) * | 2022-01-28 | 2022-05-06 | 荃豆数字科技有限公司 | Traditional Chinese medicine decoction piece refined operation guidance method and device and computer equipment |
| CN114529071A (en) * | 2022-02-11 | 2022-05-24 | 杭州致成电子科技有限公司 | Method for predicting power consumption of transformer area |
| JP7100728B1 (en) * | 2021-01-28 | 2022-07-13 | アイエムエス ソフトウェア サービシズ リミテッド | Information processing equipment, information processing methods and programs |
| CN114757703A (en) * | 2022-04-14 | 2022-07-15 | 荃豆数字科技有限公司 | Method and device for replenishing traditional Chinese medicine commodities, electronic equipment and storage medium |
| CN116051170A (en) * | 2023-01-28 | 2023-05-02 | 航天正通汇智(北京)科技股份有限公司 | Method and device for predicting regional usual medicine demand based on AI technology |
| CN116777508A (en) * | 2023-06-25 | 2023-09-19 | 急尼优医药科技(上海)有限公司 | Medical supply analysis management system and method based on big data |
-
2019
- 2019-01-17 CN CN201910044839.8A patent/CN109785003A/en active Pending
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110491120A (en) * | 2019-07-05 | 2019-11-22 | 广州车行易科技股份有限公司 | A kind of road traffic control method and its system based on big data |
| CN110619418A (en) * | 2019-07-26 | 2019-12-27 | 重庆大学 | Multi-feature water quality prediction method based on mixed model combination algorithm |
| CN110599234A (en) * | 2019-08-13 | 2019-12-20 | 广东工业大学 | Product sales prediction method |
| CN110717628A (en) * | 2019-10-09 | 2020-01-21 | 浪潮软件股份有限公司 | Goods source optimal distribution model construction method, optimal distribution model and optimal distribution method |
| CN110717628B (en) * | 2019-10-09 | 2023-05-23 | 浪潮软件股份有限公司 | Goods source optimizing model construction method, optimizing model and optimizing method |
| CN111310977A (en) * | 2020-01-20 | 2020-06-19 | 重庆亚德科技股份有限公司 | Big data-based drug inventory prediction system and method |
| CN111428930A (en) * | 2020-03-24 | 2020-07-17 | 中电药明数据科技(成都)有限公司 | GBDT-based medicine patient using number prediction method and system |
| CN111445095A (en) * | 2020-05-18 | 2020-07-24 | 江苏电力信息技术有限公司 | Annual ex-warehouse amount prediction method based on exponential smoothing method |
| CN112001740A (en) * | 2020-06-19 | 2020-11-27 | 南京理工大学 | Combined prediction method based on adaptive neural network |
| CN112053233A (en) * | 2020-09-04 | 2020-12-08 | 天元大数据信用管理有限公司 | Dynamic small and medium enterprise credit scoring method and system based on GRA |
| CN112053233B (en) * | 2020-09-04 | 2023-11-14 | 天元大数据信用管理有限公司 | GRA-based dynamic medium and small enterprise credit scoring method and system |
| CN112133415A (en) * | 2020-09-14 | 2020-12-25 | 微医云(杭州)控股有限公司 | Method, device, equipment and storage medium for replacing medicament |
| CN112148557B (en) * | 2020-09-15 | 2021-10-08 | 北京基调网络股份有限公司 | Method for predicting performance index in real time, computer equipment and storage medium |
| CN112148557A (en) * | 2020-09-15 | 2020-12-29 | 北京基调网络股份有限公司 | Method for predicting performance index in real time, computer equipment and storage medium |
| CN113190535A (en) * | 2021-01-19 | 2021-07-30 | 云南卫士盾科技有限公司 | Gas concentration prediction method based on improved grey-long and short-term memory neural network combined model |
| CN113190535B (en) * | 2021-01-19 | 2023-04-18 | 云南卫士盾科技有限公司 | Gas concentration prediction method based on improved grey-long and short-term memory neural network combined model |
| JP7100728B1 (en) * | 2021-01-28 | 2022-07-13 | アイエムエス ソフトウェア サービシズ リミテッド | Information processing equipment, information processing methods and programs |
| WO2022160370A1 (en) * | 2021-01-29 | 2022-08-04 | 南威软件股份有限公司 | Early warning method for novel coronavirus epidemic based on improved triple exponential smoothing model and lstm model |
| CN112820414B (en) * | 2021-01-29 | 2021-11-09 | 南威软件股份有限公司 | Early warning method for new crown epidemic situation based on improved cubic exponential smoothing model and LSTM model |
| CN112820414A (en) * | 2021-01-29 | 2021-05-18 | 南威软件股份有限公司 | Early warning method for new crown epidemic situation based on improved cubic exponential smoothing model and LSTM model |
| CN113361972A (en) * | 2021-07-13 | 2021-09-07 | 壹药网科技(上海)股份有限公司 | Calculation system of dominant drugs |
| CN113361972B (en) * | 2021-07-13 | 2023-09-15 | 壹药网科技(上海)股份有限公司 | System for calculating dominant drug |
| CN113837794A (en) * | 2021-08-26 | 2021-12-24 | 润联软件系统(深圳)有限公司 | Chain retail store sales prediction method based on space-time graph convolutional network |
| CN113724831A (en) * | 2021-08-31 | 2021-11-30 | 平安国际智慧城市科技股份有限公司 | Artificial intelligence-based medicine allocation method and device, terminal equipment and medium |
| CN114444946A (en) * | 2022-01-28 | 2022-05-06 | 荃豆数字科技有限公司 | Traditional Chinese medicine decoction piece refined operation guidance method and device and computer equipment |
| CN114529071A (en) * | 2022-02-11 | 2022-05-24 | 杭州致成电子科技有限公司 | Method for predicting power consumption of transformer area |
| CN114757703A (en) * | 2022-04-14 | 2022-07-15 | 荃豆数字科技有限公司 | Method and device for replenishing traditional Chinese medicine commodities, electronic equipment and storage medium |
| CN116051170A (en) * | 2023-01-28 | 2023-05-02 | 航天正通汇智(北京)科技股份有限公司 | Method and device for predicting regional usual medicine demand based on AI technology |
| CN116777508A (en) * | 2023-06-25 | 2023-09-19 | 急尼优医药科技(上海)有限公司 | Medical supply analysis management system and method based on big data |
| CN116777508B (en) * | 2023-06-25 | 2024-03-12 | 急尼优医药科技(上海)有限公司 | Medical supply analysis management system and method based on big data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785003A (en) | A kind of Pharmaceutical retail industry medicine sales forecasting system and method | |
| Lin et al. | Short-term load forecasting based on LSTM networks considering attention mechanism | |
| Hong et al. | Probabilistic electric load forecasting: A tutorial review | |
| Hajizadeh et al. | Application of data mining techniques in stock markets: A survey | |
| Marcellino et al. | Macroeconomic forecasting in the euro area: Country specific versus area-wide information | |
| Guo et al. | A multivariate intelligent decision-making model for retail sales forecasting | |
| Chen et al. | Gray relation analysis and multilayer functional link network sales forecasting model for perishable food in convenience store | |
| CN110009432A (en) | A kind of personal consumption behavior prediction technique | |
| CN102982229B (en) | The data preprocessing method of a kind of multi items price forecasting of commodity based on neural network | |
| CN111091241A (en) | BP neural network-based drug sales prediction and decision method and system | |
| Kaushik et al. | Using LSTMs for predicting patient's expenditure on medications | |
| Alamsyah et al. | Artificial neural network for predicting indonesian economic growth using macroeconomics indicators | |
| CN115796936A (en) | Cigarette sales prediction method and system based on combined model and storage medium | |
| CN115760380A (en) | Enterprise credit assessment method and system integrating electricity utilization information | |
| Pemathilake et al. | Sales forecasting based on autoregressive integrated moving average and recurrent neural network hybrid model | |
| Chen et al. | Constructing a Sales Forecasting Model by Integrating GRA and ELM: A Case Study for Retail Industry. | |
| Galkin et al. | Forecasting time series using neural networks on the example of primary sales of a pharmaceutical company | |
| Cerchiari et al. | Data mining in distribution consumer database using rough sets and self-organizing maps | |
| Zhan et al. | Small-scale demographic sequences projection based on time series clustering and lstm-rnn | |
| Khumaidi et al. | Forecasting of Sales Based on Long Short Term Memory Algorithm with Hyperparameter | |
| Jackson et al. | Machine learning for classification of economic recessions | |
| Saen | The use of artificial neural networks for technology selection in the presence of both continuous and categorical data | |
| Somaratna et al. | Which is better for inflation forecasting? Neural networks or data mining | |
| CN108470227A (en) | Based on the Regional Macro economic System prediction modeling method of assessment GDP growth figures and application | |
| Huang et al. | A novel model for stock portfolio based on ARX, RS and a new grey relational grade theories |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190521 |