CN111612535A - GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy - Google Patents

GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy Download PDF

Info

Publication number
CN111612535A
CN111612535A CN202010423918.2A CN202010423918A CN111612535A CN 111612535 A CN111612535 A CN 111612535A CN 202010423918 A CN202010423918 A CN 202010423918A CN 111612535 A CN111612535 A CN 111612535A
Authority
CN
China
Prior art keywords
model
rvm
gru
prediction
demand prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010423918.2A
Other languages
Chinese (zh)
Inventor
刘新
贺建军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202010423918.2A priority Critical patent/CN111612535A/en
Publication of CN111612535A publication Critical patent/CN111612535A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a GRU-RVM integrated model for solving the influence of sample size on the accuracy of medicine demand prediction. And respectively establishing a GRU model and an RVM model aiming at large and small sample data prediction by considering the problem that the sample size of the medicine demand prediction is uncertain. Optimizing a GRU model by using a ReLU activation function, and training the model by using an AdaMod optimizer with learning rate reduction; and optimizing the RVM model by using a mixed kernel function, and optimizing kernel function parameters and weighting coefficients by using a BAS algorithm. And then, the two models are integrated in parallel to construct a weak learner based on the Bagging idea, so that the generalization capability of the integrated models is improved. And finally, based on the AdaBoost idea, the weak learners are serially integrated to construct a hybrid integrated drug demand prediction model, so that the prediction accuracy of the integrated model is improved. According to the method, the two demand prediction models are mixed and integrated, so that the influence of sample size change on the medicine demand prediction accuracy can be eliminated to a certain extent, and the prediction precision and the generalization capability of the mixed and integrated model are effectively improved.

Description

GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy
Technical Field
The invention relates to a GRU-RVM integrated model for solving the influence of sample size on the accuracy of medicine demand prediction.
Background
The medicine has various types, and different medicines can correspondingly solve different diseases. For conventional diseases, people can select medicines according to medical orders and requirements of the people, so that the flow rates of different types of medicines and even the same type of medicines are greatly different, and the reasonable prediction of the medicine requirements plays a crucial role in good operation of pharmaceutical enterprises. At present, a conventional prediction model can only play a good prediction role for a specific sample, but when the sample changes, the conventional model often cannot play a role accurately. Particularly, for a special product, namely a medicine, the sales information of different medicines is greatly different, and when medicine demand prediction is carried out, some medicines are rich in sales information and large in sample size, but medicines with small sample size exist. When the sample size changes, the prediction model may affect the prediction effect due to under-learning or over-learning.
Disclosure of Invention
In order to solve the problem of influence of sample size on the accuracy of medicine demand prediction, the invention provides a GRU-RVM integrated model to improve the prediction precision and generalization capability of a hybrid integrated model
In order to achieve the purpose, the specific technical scheme of the invention is as follows: a GRU-RVM integration model for solving the influence of sample size on the accuracy of medicine demand prediction, comprising the following steps:
1) establishing an Adamod-GRU medicine demand prediction model aiming at large sample data prediction;
2) establishing a BAS-RVM drug demand prediction model aiming at small sample data prediction;
3) based on the Bagging idea, an Adamod-GRU model and a BAS-RVM are integrated in parallel to construct a weak learner;
4) and serially integrating weak learners based on the Adaboost idea to construct a hybrid integrated drug demand prediction model.
Further, in the step 1), establishing a prediction model of Adamod-GRU drug demand for large sample data prediction includes the following steps:
1.1) establishing a GRU medicine demand prediction model aiming at large sample data prediction;
1.2) optimizing the GRU model by utilizing a Relu activation function, and eliminating the gradient disappearance problem of the GRU model;
1.3) an AdaMod optimizer training model with learning rate reduction is used for solving the problem that a ReLU activation function is easy to cause neuron necrosis;
further, in the step 2), establishing a BAS-RVM drug demand prediction model for small sample data prediction comprises the following steps:
2.1) establishing an RVM medicine demand prediction model aiming at small sample data prediction;
2.2) optimizing the RVM model by utilizing a mixed kernel function in order to give consideration to the learning ability and the generalization ability of the RVM model;
2.3) optimizing the kernel function parameters and the weighting coefficients by using a BAS algorithm to improve the fitting degree of the model.
Further, in the step 3), the parallel integration of the Adamod-GRU model and the BAS-RVM to construct the weak learner based on the Bagging idea comprises the following steps:
3.1) respectively training an AdaMod-GRU model and a BAS-RVM model to carry out parameter optimization by taking a training set with weights D (n) as input;
3.2) predicting the AdaMod-GRU model and the BAS-RVM model after parameter optimization through a sample training set, and calculating the output h after predictionnj(x) Calculating the corresponding weight omega according to the prediction errornj. The larger the error of the prediction model is, the smaller the corresponding weight value is;
3.3) integrating the prediction results of the two models by a weighted average method, wherein the calculation formula is as follows:
Figure BDA0002497961360000021
3.4)hn(x) I.e., as the output of the weak learner at the nth level.
Further, the step 4) of serially integrating the weak learner to construct the hybrid integrated drug demand prediction model based on the Adaboost concept includes the following steps:
4.1) selecting historical information data of the drug sales volume as input of a model, and establishing a drug sales volume sample data set D;
4.2) initializing the sample set weight, and calculating the formula as follows:
Figure BDA0002497961360000022
4.3) for N ═ 1,2, …, N, training data using the training set of weights d (N) results in weak learner Hn(x) Calculating the maximum error E on the training setkExponential error e for each sampleniAnd the regression error rate enThe calculation formula is as follows:
En＝max|yi-hn(xi)|，i＝1,2,…,m
Figure BDA0002497961360000023
Figure BDA0002497961360000024
4.4) calculating the weight coefficient a of the weak learnernAnd updating the weight distribution of the sample set, ZnFor the normalization factor, the calculation formula is:
Figure BDA0002497961360000025
Figure BDA0002497961360000026
Figure BDA0002497961360000031
4.5) constructing a linear combination of the weak learners, and calculating a final strong learner H (x), wherein the calculation formula is as follows:
Figure BDA0002497961360000032
wherein g (x) is all of anhn(x) And t is 1,2, …, the median of N.
Advantageous effects
Compared with the prior art, the invention has the advantages that:
1. the invention provides an AdaMod-GRU medicine demand prediction model aiming at large sample data prediction for the first time. Firstly, eliminating the gradient disappearance problem of a GRU model by utilizing a ReLU activation function; secondly, an AdaMod optimizer with learning rate reduction is used for training and improving a GRU model for the first time, the problem that a ReLU activation function is prone to cause neuron necrosis is solved, and prediction accuracy of large sample data prediction is improved.
2. The invention provides a BAS-RVM drug demand prediction model aiming at small sample data prediction for the first time. Firstly, optimizing an RVM model by using a mixed kernel function, so that the model gives consideration to learning ability and generalization ability; and optimizing the kernel function parameters and the weighting coefficients of the improved RVM model by using a celestial cow whisker search (BAS) algorithm for the first time, so that the prediction accuracy of small sample data prediction is improved.
3. The invention firstly proposes that an AdaMod-GRU model and a BAS-RVM model are integrated in parallel through a Bagging idea, and a weak learner is constructed to improve the generalization capability of the integrated model.
4. The invention firstly proposes to combine the Bagging idea and the AdaBoost idea to construct a hybrid integrated drug demand prediction model, thereby solving the influence of sample size change on the drug demand prediction accuracy. The reliability of the prediction method of the invention is proved by experiments.
Drawings
FIG. 1 is a flow chart of a GRU-RVM integration model to account for the impact of sample size on accuracy of drug demand prediction in accordance with a first embodiment of the present invention;
FIG. 2 is a flow chart of a model for predicting the demand of Adamod-GRU drugs according to a first embodiment of the present invention;
FIG. 3 is a flow chart of a BAS-RVM drug requirement prediction model according to a first embodiment of the present invention;
FIG. 4 is a flowchart of a hybrid integrated drug demand prediction model according to a first embodiment of the present invention;
FIG. 5 is a graph comparing the convergence curves of models trained using a large sample training set according to an embodiment of the present invention;
FIG. 6 is a graph comparing the convergence curves of models trained with a small sample training set according to an embodiment of the present invention.
Detailed Description
In order to facilitate an understanding of the invention, the invention will be described more fully and in detail below with reference to the accompanying drawings and preferred embodiments, but the scope of the invention is not limited to the specific embodiments below.
The embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
Example one
Referring to fig. 1, fig. 1 is a GRU-RVM integration model for solving the influence of sample size on the accuracy of drug demand prediction according to an embodiment of the present invention, including:
step 1), establishing an Adamod-GRU medicine demand prediction model for large sample data prediction specifically comprises the following steps;
step 1.1), establishing a GRU medicine demand prediction model aiming at large sample data prediction;
step 1.2), the Relu activation function calculation formula is as follows:
Figure BDA0002497961360000041
step 1.3), training a GRU model by using an AdaMod optimizer with learning rate reduction comprises the following steps:
1.3.1) updating the exponential moving mean and the squared gradient of the gradient, β1、β2∈ [0,1) control moving average exponential decay Rate
Figure BDA0002497961360000042
mt＝β1mt-1+(1-β1)gt(3)
Figure BDA0002497961360000043
1.3.2) correcting the deviation, and reducing the influence of the deviation on the initial training
Figure BDA0002497961360000044
Figure BDA0002497961360000045
1.3.3) to solve the abnormal over-value of the adaptive learning rate η and avoid the optimizer from being in a bad state, a hyper-parameter β is introduced3As a measure of the length of memory, the closer it is to 1, the longer the length of memory
Figure BDA0002497961360000046
st＝β3st-1+(1-β3)nt(8)
Figure BDA0002497961360000047
1.3.4) performing a gradient update
Figure BDA0002497961360000048
1.3.5) judging whether the condition of iteration ending is reached, if so, outputting the optimizing parameter value, otherwise, repeating 1.3.1) to 1.3.4) until the condition is met.
The Adamod-GRU model flow is shown in fig. 2.
Step 2), establishing a BAS-RVM drug demand prediction model aiming at small sample data prediction specifically comprises the following steps:
step 2.2), establishing an RVM medicine demand prediction model aiming at small sample data prediction;
step 2.2), the construction of the mixed kernel function comprises the following steps:
2.2.1) adopting a polynomial kernel function with stronger global property to ensure the sufficient learning of the training data, and improving the learning ability of the RVM model. The polynomial kernel function calculation formula is:
k1(x,y)＝(αxTy+c)d(11)
2.2.2) adopting a Gaussian kernel function with stronger locality to avoid the interference of noise in training data, and improving the generalization capability of the RVM model. The gaussian kernel function calculation formula is:
k2(x,y)＝exp(-‖x-y‖2/2σ2) (12)
2.2.3) constructing a mixed kernel function by a weighted average method, wherein the calculation formula is as follows:
k(x,y)＝λk1(x,y)+(1-λ)k2(x,y) (13)
step 2.3), the optimization of the kernel function parameters and the weighting coefficients by the BAS algorithm comprises the following steps:
2.3.1) establishing a fitness function f (x), x ═ x1,x2,…,xn]TWherein x represents an n-dimensional vector composed of parameters of all sub-kernel functions in the mixed kernel function and weights thereof;
2.3.2) the direction of pointing the right whisker to the left whisker of the longicorn is the direction of the longicorn, theoretically, the direction is random at this time, and the direction is normalized to a unit vector, and the calculation formula is as follows:
dir＝rand(n,1)/‖rand(n,1)‖ (14)
where rand () represents a random vector and n represents a spatial dimension.
2.3.3) establishing the position coordinates of the left and right longicorn whiskers, wherein the calculation formula is as follows:
Figure BDA0002497961360000051
in the formula, xlRepresenting the left whisker coordinate, xrRepresenting the coordinates of the right whisker, x the coordinates of the center of mass, d0The distance between the left and right whiskers.
2.3.4) calculating the fitness function value f (x) of the left and right whiskersl) And f (x)r) And judging the magnitude of the two values. To find the fitness function minimum, if f (x)l)<f(xr) Then the longicorn must move one step toward the left beard, if f (x)l)<f(xr) If the longhorn must move one step towards the left beard direction, the calculation formula is:
x＝x+s×dir×sign(f(xl)-f(xr)) (16)
where s denotes the step size per move and sign is a sign function.
2.3.5) judging whether the condition of iteration ending is reached, if the condition of iteration ending is met, outputting the optimizing parameter value, otherwise, repeating step2 to step4 until the condition is met.
The BAS-RVM model flow is shown in FIG. 3.
Step 3), the parallel integration of the Adamod-GRU model and the BAS-RVM to construct the weak learner based on the Bagging idea specifically comprises the following steps:
step 3.1), using the training set with weight D (n) as input to train AdaMod-GRU model and BAS-
Performing parameter optimization on the RVM model;
step 3.2), predicting the AdaMod-GRU model and the BAS-RVM model after parameter optimization through a sample training set, and calculating the predicted output hnj(x) Calculating the corresponding weight omega according to the prediction errornj. The larger the error of the prediction model is, the smaller the corresponding weight value is;
step 3.3), integrating the prediction results of the two models by a weighted average method, wherein the calculation formula is as follows:
Figure BDA0002497961360000061
step 3.4) hn(x) I.e., as the output of the weak learner at the nth level.
Step 4), serially integrating the weak learners based on the Adaboost idea to construct a hybrid integrated drug demand prediction model specifically comprises the following steps:
step 4.1) selecting historical information data of the drug sales volume as input of a model, and establishing a drug sales volume sample data set D;
step 4.2) initializing sample set weight, wherein the calculation formula is as follows:
Figure BDA0002497961360000062
step 4.3) training data using the training set with weight d (N) for N1, 2, …, N, to obtain weak learner Hn(x) Calculating the maximum error E on the training setkExponential error e for each sampleniAnd the regression error rate enThe calculation formula is as follows:
En＝max|yi-hn(xi)|，i＝1,2,…,m (19)
Figure BDA0002497961360000063
Figure BDA0002497961360000064
step 4.4) calculating the weight coefficient a of the weak learnernAnd updating the weight distribution of the sample set, ZnFor the normalization factor, the calculation formula is:
Figure BDA0002497961360000065
Figure BDA0002497961360000066
Figure BDA0002497961360000067
and 4.5) constructing a linear combination of the weak learners, and calculating the final strong learner H (x), wherein the calculation formula is as follows:
Figure BDA0002497961360000068
wherein g (x) is all of anhn(x) And t is 1,2, …, the median of N.
The hybrid integrated model structure is shown in fig. 4.
According to the method, the medicine sales information of a certain medicine enterprise is used as original data, the large-sample medicine sales data set and the small-sample medicine sales data set are respectively established to compare the prediction accuracy and the generalization capability of three medicine demand prediction models, and the experimental results are shown in fig. 5 and fig. 6.
Firstly, the prediction precision is compared with the experimental analysis. It can be found from the convergence curves of the three models in fig. 5: when large sample data is used for training, the correlation vector of the RVM increases with the increase of training samples, and a local convergence phenomenon occurs, so that a prediction result has a large error, and therefore, the BAS-RVM model is not suitable for predicting the large sample data. The AdaMod-GRU model and the integrated prediction model show good effect on prediction of a large sample, and compared with the AdaMod-GRU model, the integrated model is optimized in convergence speed and fitting goodness; it can be found from the convergence curves of the three models in fig. 6 that: when small sample data is adopted for training, the GRU cannot be converged due to lack of learning samples, so that a prediction result has a large error, and AdaMod-GRU is not suitable for predicting the small sample data. The BAS-RVM model and the integrated prediction model show good effect on the prediction of small samples, and compared with the BAS-RVM model, the integrated model is optimized in convergence speed and goodness of fit. The comparison experiment shows that the convergence speed and the convergence precision of the integrated prediction model are better than those of the model before integration, which shows that the integrated model can improve the prediction precision of the medicine demand.
Then, the generalization ability comparison experiment analysis is carried out. According to the analysis content, the prediction precision of the AdaMod-GRU model and the BAS-RVM model is greatly related to the number of training samples. The integrated prediction model is reversely observed, and when the data samples change, the prediction capability of the model is not greatly changed, which shows that the integrated model has certain generalization capability.
Combining the results of two comparative experiments, it can be found that: the integrated prediction model has the advantage that the generalization capability of the model is enhanced due to the existence of a weak learner which is constructed by AdaMod-GRU and BAS-RVM models based on Bagging thought. Meanwhile, the weak learner is subjected to weight accumulation to form the strong learner based on the AdaBoost idea, so that the prediction precision of the model is also remarkably improved. Theoretical analysis and experimental verification prove that the integrated medicine demand prediction model based on classification learning has good prediction accuracy and generalization capability.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes will occur to those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (5)

1. A GRU-RVM integration model for solving the influence of sample size on the accuracy of medicine demand prediction is characterized in that: the method comprises the following steps:
1) establishing an Adamod-GRU medicine demand prediction model aiming at large sample data prediction;
2) establishing a BAS-RVM drug demand prediction model aiming at small sample data prediction;
3) based on the Bagging idea, an Adamod-GRU model and a BAS-RVM are integrated in parallel to construct a weak learner;
4) and serially integrating weak learners based on the Adaboost idea to construct a hybrid integrated drug demand prediction model.
2. The GRU-RVM integration model of claim 1 that accounts for sample size impact on accuracy of drug demand forecasting, characterized by: in the step 1), the establishment of an Adamod-GRU medicine demand prediction model for large sample data prediction comprises the following steps:
1.1) establishing a GRU medicine demand prediction model aiming at large sample data prediction;
1.2) optimizing the GRU model by utilizing a Relu activation function, and eliminating the gradient disappearance problem of the GRU model;
1.3) training a model by using an AdaMod optimizer with learning rate reduction, and solving the problem that the ReLU activation function is easy to cause neuron necrosis.
3. The GRU-RVM integration model of claim 1 that accounts for sample size impact on accuracy of drug demand forecasting, characterized by: in the step 2), the establishment of the BAS-RVM drug demand prediction model aiming at small sample data prediction comprises the following steps:
2.1) establishing an RVM medicine demand prediction model aiming at small sample data prediction;
2.2) optimizing the RVM model by utilizing a mixed kernel function in order to give consideration to the learning ability and the generalization ability of the RVM model;
2.3) optimizing the kernel function parameters and the weighting coefficients by using a BAS algorithm to improve the fitting degree of the model.
4. The GRU-RVM integration model of claim 1 that accounts for sample size impact on accuracy of drug demand forecasting, characterized by: in the step 3), the parallel integration construction of the Adamod-GRU model and the BAS-RVM weak learner based on the Bagging thought comprises the following steps:
3.1) respectively training an AdaMod-GRU model and a BAS-RVM model to carry out parameter optimization by taking a training set with weights D (n) as input;
3.2) predicting the AdaMod-GRU model and the BAS-RVM model after parameter optimization through a sample training set, and calculating the output h after predictionnj(x) Calculating the corresponding weight omega according to the prediction errornj(ii) a The larger the error of the prediction model is, the smaller the corresponding weight value is;
3.3) integrating the prediction results of the two models by a weighted average method, wherein the calculation formula is as follows:
Figure FDA0002497961350000011
3.4)hn(x) I.e., as the output of the weak learner at the nth level.
5. The GRU-RVM integration model of claim 1 that accounts for sample size impact on accuracy of drug demand forecasting, characterized by: in the step 4), the method for serially integrating the weak learner to construct the hybrid integrated drug demand prediction model based on the Adaboost idea comprises the following steps:
4.1) selecting historical information data of the drug sales volume as input of a model, and establishing a drug sales volume sample data set D;
4.2) initializing the sample set weight, and calculating the formula as follows:
Figure FDA0002497961350000021
4.3) for N ═ 1,2, …, N, training data using the training set of weights d (N) results in weak learner Hn(x) Calculating the maximum error E on the training setkExponential error e for each sampleniAnd the regression error rate enThe calculation formula is as follows:
En＝max|yi-hn(xi)|，i＝1,2,…,m
Figure FDA0002497961350000022
Figure FDA0002497961350000023
4.4) calculating the weight coefficient a of the weak learnernAnd updating the weight distribution of the sample set, ZnFor the normalization factor, the calculation formula is:
Figure FDA0002497961350000024
Figure FDA0002497961350000025
Figure FDA0002497961350000026
4.5) constructing a linear combination of the weak learners, and calculating a final strong learner H (x), wherein the calculation formula is as follows:
Figure FDA0002497961350000027
wherein g (x) is all of anhn(x) And t is 1,2, …, the median of N.
CN202010423918.2A 2020-05-19 2020-05-19 GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy Pending CN111612535A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010423918.2A CN111612535A (en) 2020-05-19 2020-05-19 GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010423918.2A CN111612535A (en) 2020-05-19 2020-05-19 GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy

Publications (1)

Publication Number Publication Date
CN111612535A true CN111612535A (en) 2020-09-01

Family

ID=72202104

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010423918.2A Pending CN111612535A (en) 2020-05-19 2020-05-19 GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy

Country Status (1)

Country Link
CN (1) CN111612535A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215412A (en) * 2020-09-27 2021-01-12 中国农业大学 Dissolved oxygen prediction method and device
CN114049162A (en) * 2022-01-11 2022-02-15 北京京东振世信息技术有限公司 Model training method, demand prediction method, apparatus, device, and storage medium

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215412A (en) * 2020-09-27 2021-01-12 中国农业大学 Dissolved oxygen prediction method and device
CN112215412B (en) * 2020-09-27 2023-12-22 中国农业大学 Dissolved oxygen prediction method and device
CN114049162A (en) * 2022-01-11 2022-02-15 北京京东振世信息技术有限公司 Model training method, demand prediction method, apparatus, device, and storage medium

Similar Documents

Publication Publication Date Title
Tzanakou Supervised and unsupervised pattern recognition: feature extraction and computational intelligence
Wahba et al. Soft classification, aka risk estimation, via penalized log likelihood and smoothing spline analysis of variance
US11023806B2 (en) Learning apparatus, identifying apparatus, learning and identifying system, and recording medium
US20210256392A1 (en) Automating the design of neural networks for anomaly detection
US11593611B2 (en) Neural network cooperation
Alejo et al. A hybrid method to face class overlap and class imbalance on neural networks and multi-class scenarios
Burbidge et al. Drug design by machine learning: support vector machines for pharmaceutical data analysis
Yap et al. An application of oversampling, undersampling, bagging and boosting in handling imbalanced datasets
US20190236482A1 (en) Training machine learning models on multiple machine learning tasks
Lughofer et al. On employing fuzzy modeling algorithms for the valuation of residential premises
Stinis et al. Enforcing constraints for interpolation and extrapolation in generative adversarial networks
Elgamal et al. Improved reptile search optimization algorithm using chaotic map and simulated annealing for feature selection in medical field
CN111612535A (en) GRU-RVM integrated model for solving influence of sample size on medicine demand prediction accuracy
Zahavy et al. Deep neural linear bandits: Overcoming catastrophic forgetting through likelihood matching
Chao et al. A fuzzy adaptive controller for cuckoo search algorithm in active suspension system
AU2022424925A1 (en) Processing sequences of multi-modal entity features using convolutional neural networks
Xie et al. Modeling adaptive preview time of driver model for intelligent vehicles based on deep learning
Pauls et al. Determining optimum drop-out rate for neural networks
Le-Duc et al. Strengthening gradient descent by sequential motion optimization for deep neural networks
Hardy et al. Multi-step prediction of nonlinear Gaussian Process dynamics models with adaptive Gaussian mixtures
Jordanov et al. Neural network learning with global heuristic search
Geng et al. Design of autonomous vehicle trajectory tracking controller based on Neural Network Predictive Control
Lyu et al. Efficient factorisation-based Gaussian process approaches for online tracking
CN112257861A (en) Double-layer learning model for pattern recognition and classification, construction method and application
Briffoteaux Parallel surrogate-based algorithms for solving expensive optimization problems

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200901

WD01 Invention patent application deemed withdrawn after publication