CN111276229A - Outpatient quantity prediction method and system based on deep belief network - Google Patents

Outpatient quantity prediction method and system based on deep belief network Download PDF

Info

Publication number
CN111276229A
CN111276229A CN202010110803.8A CN202010110803A CN111276229A CN 111276229 A CN111276229 A CN 111276229A CN 202010110803 A CN202010110803 A CN 202010110803A CN 111276229 A CN111276229 A CN 111276229A
Authority
CN
China
Prior art keywords
layer
network
data
prediction
outpatient
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
CN202010110803.8A
Other languages
Chinese (zh)
Inventor
谷兴龙
李向阳
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.)
Shandong Health Medical Big Data Co ltd
Original Assignee
Shandong Health Medical Big Data Co ltd
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 Shandong Health Medical Big Data Co ltd filed Critical Shandong Health Medical Big Data Co ltd
Priority to CN202010110803.8A priority Critical patent/CN111276229A/en
Publication of CN111276229A publication Critical patent/CN111276229A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
    • G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Abstract

The invention discloses an outpatient quantity prediction method and system based on a deep belief network, belonging to the technical field of data mining analysis, aiming at solving the technical problem of more effectively and accurately predicting the outpatient quantity of a hospital according to the historical outpatient quantity, and the technical scheme is as follows: the method is based on a restricted Boltzmann machine and a mode of establishing a logistic regression layer, and utilizes historical outpatient data of a hospital to predict daily outpatient quantity, weekly outpatient quantity and monthly outpatient quantity, so that the outpatient quantity of the hospital at the current time can be predicted more accurately and effectively; the method comprises the following specific steps: s1, constructing a deep confidence network; s2, establishing a logistic regression layer: adding a logistic regression layer on the upper layer of the deep confidence network as a prediction layer, and performing supervised prediction on the clinic diagnosis amount data by combining the extracted data characteristics of the historical clinic diagnosis amount; s3, training the network: training of the neural network is performed using a greedy layer-by-layer algorithm. The invention also discloses an outpatient quantity prediction system based on the deep belief network.

Description

Outpatient quantity prediction method and system based on deep belief network
Technical Field
The invention relates to the technical field of data mining analysis, in particular to an outpatient quantity prediction method and system based on a deep belief network.
Background
The outpatient service of the hospital is the first link of the patients receiving the treatment work of the hospital, is a window of the hospital facing the society, and is favorable for the smooth expansion of the subsequent medical work by accurately predicting the outpatient service quantity of the hospital. The hospital outpatient quantity can be used as an important index for evaluating the medical working condition of the modern hospital, and is also beneficial to the hospital to have important reference value for medical resources and distribution, so that the effective method is used for carrying out feature extraction on the historical outpatient quantity data of the hospital, and is further used for predicting the hospital outpatient quantity, and more attention is paid.
The prediction method for the hospital clinic volume is multiple, the basic idea is to analyze the relation and the rule among the historical clinic volume data and predict the clinic volume in a certain time period in the future, and the main methods can be roughly divided into three categories:
(1) and a time series analysis model based on historical data: a representative method is an autoregressive integrated moving average model (ARIMA), which analyzes the rule of time variation of hospital clinic volume so as to predict hospital clinic volume, and the model has better performance in predicting monthly clinic volume and quarterly clinic volume, but the ARIMA model assumes that data are linear in advance, and if the data have a nonlinear relationship, the performance of the model is reduced;
(2) and an artificial intelligence-based model: representative methods include Artificial Neural Networks (ANNs), Support Vector Machines (SVM), Genetic Algorithms (GA), and the like, and the methods can better simulate the nonlinear relationship existing among hospital outpatient data.
(3) The time series analysis model based on historical data and the model based on artificial intelligence are mixed models of two methods: the hybrid model combines the two methods to predict the hospital clinic amount, on one hand, the advantages of the different methods are utilized, and the defects of the two methods are avoided, so that the prediction result of the general hybrid model is more accurate compared with that of a single method, but the method is complex to operate.
In summary, how to more effectively and accurately predict the outpatient quantity of a hospital according to the historical outpatient quantity is an urgent problem to be solved at present
Disclosure of Invention
The technical task of the invention is to provide a clinic volume prediction method and a clinic volume prediction system based on a deep belief network, so as to solve the problem of how to more effectively and accurately predict the clinic volume of a hospital according to the historical clinic volume.
The technical task of the invention is realized in the following way, the method for predicting the number of outpatients based on the deep confidence network is based on a Restricted Boltzmann Machine (RBM) and a way of establishing a logistic regression layer, and the method utilizes the historical outpatient data of a hospital to predict the number of daily outpatients, the number of weekly outpatients and the number of monthly outpatients, so as to realize more accurate and effective prediction of the number of outpatients of the hospital at the current time; the method comprises the following specific steps:
s1, constructing a Deep Belief Network (DBN): forming a deep confidence network (DBN) based on a plurality of limited Boltzmann machines (RBMs), wherein the output trained by the limited Boltzmann machine (RBM) of each layer is used as the input of the limited Boltzmann machine (RBM) of the next layer to form the deep confidence network (DBN), and the whole deep confidence network (DBN) is used as a data feature extraction layer for extracting the data features of the historical clinic amount;
s2, establishing a logistic regression layer: a logistic regression layer is added on the upper layer of a Deep Belief Network (DBN) to serve as a prediction layer, and the clinic volume data is subjected to supervised prediction by combining the extracted data characteristics of the historical clinic volume, so that the prediction precision is improved;
s3, training the network: training of the neural network is performed using a greedy layer-by-layer algorithm.
Preferably, the Deep Belief Network (DBN) is constructed in step S1 as follows:
s101, the lowest layer of the depth confidence network is a Restricted Boltzmann Machine (RBM), and the Restricted Boltzmann Machine (RBM) consists of a visible layer (data input layer) and a hidden layer (feature extraction layer);
s102, setting the number of neurons in a visible layer to be 6 (the number can be adjusted according to actual conditions), setting the number of neurons in a hidden layer to be 3, and setting the number of layers in the hidden layer to be 3;
s103, taking the output of a Restricted Boltzmann Machine (RBM) formed by the first visible layer and the first hidden layer as the input of the next hidden layer, thereby obtaining the output of the next hidden layer, and sequentially iterating to construct a finished depth confidence network.
Preferably, the logistic regression layer in step S2 is a regression prediction layer superimposed on the deep confidence network, and supervised prediction is performed on the hospital outpatient quantity by using the data features extracted by the deep confidence network and the selected label data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to a time interval; the method comprises the following specific steps:
s201, initializing W, b, wherein W is initialized to a number close to 0, and b is initialized to 0; wherein W represents a weight matrix;
s202, calculating the predicted output result
Figure BDA0002389924110000031
S20201, calculating data characteristics Z, wherein the formula is as follows:
Z＝WTX+b；
wherein X represents the number of samples and W represents a weight matrix; b is a constant;
s20202, calculating a result matrix A of prediction, wherein a formula is as follows:
Figure BDA0002389924110000032
wherein Y represents an actual value; x, A, Z are each vectorized matrices of samples;
s203, calculating a loss function according to a formula, wherein the formula is as follows:
Figure BDA0002389924110000033
s204, calculating a gradient, namely, performing derivation on the loss function in the step S203;
s205, updating W, b, and repeating steps S202 to S205 within the number of iterations until the derivative results in the minimized cost function J (w, b).
Preferably, the training network in step S3 is specifically as follows:
s301, training a network layer by using label-free data;
s302, after the pre-training is completed, adjusting and optimizing the parameters from top to bottom by using a BP algorithm.
Preferably, in the step S301, a contrast divergence algorithm (CD-k) is used to train the network in a layer-by-layer training network using unlabeled data, specifically as follows:
s30101, sample data x is assigned to the visible layer according to a formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function;
s30102, extracting a sample h from the calculated probability distribution by adopting Gibbs sampling1～P(h1|v1)；
S30103, use1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1) (ii) a Wherein, sigma also represents sigmoid activation function;
s30104, similarly, a Gibbs sample is taken from the probability distribution obtained by calculation to extract a sample v2～P(v2|h1)；
S30105, passing through v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
S30106, update the weight, W ← W + λ (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
Preferably, the daily out-patient volume prediction is specifically as follows:
(1) the input data is the outpatient service volume of each day in a certain selected time period, and the features among the outpatient service volume data are extracted through the constructed deep belief network, namely the input original data are converted into a group of feature vectors;
(2) after the characteristic vector is obtained, predicting the clinic diagnosis amount of the next day by utilizing a logistic regression layer and combining the characteristic vector, thereby obtaining a predicted value;
(3) comparing the label data (namely the actual value of the clinic volume data in the future day) with the predicted value to obtain the difference between the label data and the predicted value, finely adjusting the parameters through back propagation, and obtaining the optimal group of parameters after limited iteration to realize the prediction of the clinic volume in a certain period of time in the future.
An outpatient quantity prediction system based on a deep belief network, the system comprising,
the depth confidence network (DBN) construction unit is used for forming a depth confidence network (DBN) based on a plurality of limited Boltzmann machines (RBMs), the output trained by the limited Boltzmann machines (RBMs) of each layer is used as the input of the limited Boltzmann machines (RBMs) of the next layer to form the depth confidence network (DBN), and the whole depth confidence network (DBN) is used as a data feature extraction layer for extracting the data features of the historical clinic quantity;
establishing a logistic regression layer unit for adding a logistic regression layer on the upper layer of a Deep Belief Network (DBN) as a prediction layer, and performing supervised prediction on the clinic volume data by combining the extracted data characteristics of the historical clinic volume to improve the prediction precision;
and the network training unit is used for training the neural network by using a greedy layer-by-layer algorithm.
Preferably, the lowest layer of the deep confidence network (DBN) is a Restricted Boltzmann Machine (RBM) which is composed of a visible layer (data input layer) and a hidden layer (feature extraction layer); the number of the neurons in the visible layer is set to be 6 (can be adjusted according to actual conditions), the number of the neurons in the hidden layer is set to be 3, and the number of the layers in the hidden layer is set to be 3.
Preferably, the logistic regression layer is a regression prediction layer superposed on the deep confidence network, and the supervised prediction is carried out on the hospital clinic volume by using the data features extracted by the deep confidence network and the selected label data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to the time interval.
Preferably, the network training unit comprises,
the data layer-by-layer training network module is used for training a network layer by using unlabeled data;
and the tuning module is used for tuning the parameters from top to bottom by using a BP algorithm after the pre-training is finished.
The data layer-by-layer training network module trains the network by using a contrast divergence algorithm (CD-k), and the specific process is as follows:
①, sample data x is assigned to the visible layer according to the formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function;
②, extracting a sample h from the calculated probability distribution by Gibbs sampling1～P(h1|v1)；
③, use h1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1) (ii) a Wherein, sigma also represents sigmoid activation function;
④, similarly, a sample v is extracted from the computed probability distribution by Gibbs sampling2～P(v2|h1)；
⑤, passing v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
⑥, update weight, W ← W + lambda (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
The outpatient quantity prediction method and the outpatient quantity prediction system based on the deep belief network have the following advantages that:
the method has the advantages that the characteristics in the hospital historical clinic amount data are automatically extracted based on the neural network, the logistic regression layer is additionally arranged on the top layer, the clinic amount is supervised and predicted, the hospital clinic amount can be more effectively and accurately predicted in a certain time period in the future, the characteristics contained in the data are manually analyzed, a great deal of energy consumed by extracting the characteristics is avoided, and the universality of the use in various hospitals is improved;
the invention is based on a RBM, combines a plurality of hidden layers and constructs a deep confidence network, namely a feature extraction layer, which can effectively simulate the nonlinear relation among hospital outpatient data and realize the feature extraction of the data;
thirdly, a logistic regression layer is added on the top layer of the deep confidence network, so that supervised outpatient quantity prediction can be performed by utilizing label data, and the linear relation among hospital outpatient quantity data is effectively used for supervised regression prediction;
and after the pre-training of the deep network, relevant parameters of the whole network are optimized, then the BP algorithm is used for fine adjustment, and only one local search needs to be carried out on the known parameters, so that the training efficiency of the neural network is improved.
Drawings
The invention is further described below with reference to the accompanying drawings.
FIG. 1 is a flow chart of an outpatient quantity prediction method based on a deep belief network;
FIG. 2 is a schematic diagram of a basic structure of a deep belief network;
FIG. 3 is a basic block diagram of a Restricted Boltzmann Machine (RBM)
Detailed Description
The method and system for predicting out-patient quantity based on deep belief network of the present invention will be described in detail with reference to the drawings and the detailed description.
Example 1:
as shown in fig. 1, the outpatient quantity prediction method based on the deep belief network of the present invention is based on a Restricted Boltzmann Machine (RBM) and a manner of establishing a logistic regression layer, and predicts daily outpatient quantity, weekly outpatient quantity and monthly outpatient quantity by using historical outpatient data of a hospital, so as to realize more accurate and effective prediction of outpatient quantity of the hospital at the current time; the method comprises the following specific steps:
s1, constructing a Deep Belief Network (DBN): forming a deep confidence network (DBN) based on a plurality of limited Boltzmann machines (RBMs), wherein the output trained by the limited Boltzmann machine (RBM) of each layer is used as the input of the limited Boltzmann machine (RBM) of the next layer to form the deep confidence network (DBN), and the whole deep confidence network (DBN) is used as a data feature extraction layer for extracting the data features of the historical clinic amount; the method comprises the following specific steps:
s101, the lowest layer of the depth confidence network is a Restricted Boltzmann Machine (RBM), and the Restricted Boltzmann Machine (RBM) consists of a visible layer (data input layer) and a hidden layer (feature extraction layer);
s102, setting the number of neurons in a visible layer to be 6 (the number can be adjusted according to actual conditions), setting the number of neurons in a hidden layer to be 3, and setting the number of layers in the hidden layer to be 3;
s103, taking the output of a Restricted Boltzmann Machine (RBM) formed by the first visible layer and the first hidden layer as the input of the next hidden layer, thereby obtaining the output of the next hidden layer, and sequentially iterating to construct a finished depth confidence network.
S2, establishing a logistic regression layer: a logistic regression layer is added on the upper layer of a Deep Belief Network (DBN) to serve as a prediction layer, and the clinic volume data is subjected to supervised prediction by combining the extracted data characteristics of the historical clinic volume, so that the prediction precision is improved; the logistic regression layer is a regression prediction layer superposed on the deep confidence network, and supervised prediction is carried out on the hospital outpatient quantity by utilizing the data characteristics extracted by the deep confidence network and the selected label data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to a time interval; for example, if the selected time interval is 6, the number of neurons to be output is finally predicted to be 1 by using the outpatient volume of the first 6 days (weeks, months) as an input and the outpatient volume of the 7 th day (weeks, months) as label data.
The method comprises the following specific steps:
s201, initializing W, b, wherein W is initialized to a number close to 0, and b is initialized to 0; wherein W represents a weight matrix;
s202, calculating the predicted output result
Figure BDA0002389924110000071
S20201, calculating data characteristics Z, wherein the formula is as follows:
Z＝WTX+b；
wherein X represents the number of samples and W represents a weight matrix; b is a constant;
s20202, calculating a result matrix A of prediction, wherein a formula is as follows:
Figure BDA0002389924110000081
wherein Y represents an actual value; x, A, Z are each vectorized matrices of samples;
s203, calculating a loss function according to a formula, wherein the formula is as follows:
Figure BDA0002389924110000082
s204, calculating a gradient, namely, performing derivation on the loss function in the step S203;
s205, updating W, b, and repeating steps S202 to S205 within the number of iterations until the derivative results in the minimized cost function J (w, b).
S3, training the network: training a neural network by using a greedy layer-by-layer algorithm; the method comprises the following specific steps:
s301, training a network layer by using label-free data; the high-level feature representation y generated from the feature representation x from bottom to top and the feature representation x' generated from the high-level feature representation y from top to bottom are consistent as much as possible, and for a sample data x, a contrast divergence algorithm (CD-k) is adopted to train the network, which is specifically as follows:
s30101, sample data x is assigned to the visible layer according to a formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function;
s30102, extracting a sample h from the calculated probability distribution by adopting Gibbs sampling1～P(h1|v1)；
S30103, use1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1) (ii) a Wherein, sigma also represents sigmoid activation function;
s30104, similarly, a Gibbs sample is taken from the probability distribution obtained by calculation to extract a sample v2～P(v2|h1)；
S30105, passing through v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
S30106, update the weight, W ← W + λ (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
S302, after the pre-training is completed, adjusting and optimizing the parameters from top to bottom by using a BP algorithm (back propagation algorithm).
Because a plurality of RBMs are connected in series, a DBN is formed, wherein the hidden layer of the previous RBM is used as the display layer of the next RBM, and the output of the previous RBM is the input of the next RBM. In the training process, the RBMs of the previous layer are required to be fully trained, and then the RBMs of the current layer can be trained until the final layer, namely the essence of the greedy layer-by-layer training algorithm.
The specific prediction of the daily outpatient quantity is as follows:
(1) the input data is the outpatient service volume of each day in a certain selected time period, and the features among the outpatient service volume data are extracted through the constructed deep belief network, namely the input original data are converted into a group of feature vectors;
(2) after the characteristic vector is obtained, predicting the clinic diagnosis amount of the next day by utilizing a logistic regression layer and combining the characteristic vector, thereby obtaining a predicted value;
(3) comparing the label data (namely the actual value of the clinic volume data in the future day) with the predicted value to obtain the difference between the label data and the predicted value, finely adjusting the parameters through back propagation, and obtaining the optimal group of parameters after limited iteration to realize the prediction of the clinic volume in a certain period of time in the future.
Example 2:
the invention relates to a clinic quantity forecasting system based on a deep belief network, which comprises,
the depth confidence network (DBN) construction unit is used for forming a depth confidence network (DBN) based on a plurality of limited Boltzmann machines (RBMs), the output trained by the limited Boltzmann machines (RBMs) of each layer is used as the input of the limited Boltzmann machines (RBMs) of the next layer to form the depth confidence network (DBN), and the whole depth confidence network (DBN) is used as a data feature extraction layer for extracting the data features of the historical clinic quantity; the bottom layer of the deep confidence network (DBN) is a Restricted Boltzmann Machine (RBM), and the Restricted Boltzmann Machine (RBM) consists of a visible layer (data input layer) and a hidden layer (feature extraction layer); the number of the neurons in the visible layer is set to be 6 (can be adjusted according to actual conditions), the number of the neurons in the hidden layer is set to be 3, and the number of the layers in the hidden layer is set to be 3.
Establishing a logistic regression layer unit for adding a logistic regression layer on the upper layer of a Deep Belief Network (DBN) as a prediction layer, and performing supervised prediction on the clinic volume data by combining the extracted data characteristics of the historical clinic volume to improve the prediction precision; the logistic regression layer is a regression prediction layer superposed on the deep confidence network, and supervised prediction is carried out on the hospital clinic quantity by using the data characteristics extracted by the deep confidence network and the selected label data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to the time interval.
The network training unit is used for training the neural network by using a greedy layer-by-layer algorithm; the network training unit comprises a network training unit,
the data layer-by-layer training network module is used for training a network layer by using unlabeled data;
and the tuning module is used for tuning the parameters from top to bottom by using a BP algorithm after the pre-training is finished.
Wherein, the high-level feature representation y generated from the feature representation x from bottom to top and the feature representation x' generated from the high-level feature representation y from top to bottom are consistent as much as possible, and for a sample data x, the data layer-by-layer training network module trains the network by using a contrast divergence algorithm (CD-k), and the specific process is as follows:
①, sample data x is assigned to the visible layer according to the formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function;
②, extracting a sample h from the calculated probability distribution by Gibbs sampling1～P(h1|v1)；
③, use h1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1) (ii) a Wherein, sigma also represents sigmoid activation function;
④, similarly, a sample v is extracted from the computed probability distribution by Gibbs sampling2～P(v2|h1)；
⑤, passing v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
⑥, update weight, W ← W + lambda (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. The out-patient quantity prediction method based on the deep belief network is characterized in that the method is based on a limited Boltzmann machine and a mode of establishing a logistic regression layer, and the historical out-patient data of a hospital is utilized to predict the daily out-patient quantity, the weekly out-patient quantity and the monthly out-patient quantity, so that the out-patient quantity of the hospital at the current time can be more accurately and effectively predicted; the method comprises the following specific steps:
s1, constructing a deep confidence network: forming a depth confidence network based on a plurality of limited Boltzmann machines, wherein the output trained by the limited Boltzmann machine of each layer is used as the input of the limited Boltzmann machine of the next layer to form the depth confidence network, and the whole depth confidence network is used as a data feature extraction layer for extracting the data features of the historical clinic amount;
s2, establishing a logistic regression layer: a logistic regression layer is added on the upper layer of the deep confidence network to serve as a prediction layer, and the clinic volume data is subjected to supervised prediction by combining the extracted data characteristics of the historical clinic volume, so that the prediction precision is improved;
s3, training the network: training of the neural network is performed using a greedy layer-by-layer algorithm.
2. The out-patient prediction method based on the deep belief network as claimed in claim 1, wherein the deep belief network is constructed in the step S1 as follows:
s101, a limited Boltzmann machine is arranged at the bottom layer of the depth confidence network, and consists of a visible layer and a hidden layer;
s102, setting the number of neurons in a visible layer to be 6, setting the number of neurons in a hidden layer to be 3, and setting the number of layers in the hidden layer to be 3;
s103, taking the output of the restricted Boltzmann machine consisting of the first visible layer and the first hidden layer as the input of the next hidden layer, so as to obtain the output of the next hidden layer, and sequentially iterating to construct a finished depth confidence network.
3. The method for outpatient quantity prediction based on deep belief network as claimed in claim 1, wherein the logistic regression layer in step S2 is a regression prediction layer superimposed on the deep belief network, and the hospital outpatient quantity is supervised predicted by using the data features extracted by the deep belief network and the selected tag data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to a time interval; the method comprises the following specific steps:
s201, initializing W, b, wherein W is initialized to a number close to 0, and b is initialized to 0; wherein W represents a weight matrix;
s202, calculating the predicted output result
Figure FDA0002389924100000021
S20201, calculating data characteristics Z, wherein the formula is as follows:
Z＝WTX+b；
wherein X represents the number of samples and W represents a weight matrix; b is a constant;
s20202, calculating a result matrix A of prediction, wherein a formula is as follows:
Figure FDA0002389924100000022
wherein Y represents an actual value; x, A, Z are each vectorized matrices of samples;
s203, calculating a loss function according to a formula, wherein the formula is as follows:
Figure FDA0002389924100000023
s204, calculating a gradient, namely, performing derivation on the loss function in the step S203;
s205, updating W, b, and repeating steps S202 to S205 within the number of iterations until the derivative results in the minimized cost function J (w, b).
4. The out-patient prediction method based on deep belief network as claimed in claim 1, wherein the training network in step S3 is specifically as follows:
s301, training a network layer by using label-free data;
s302, after the pre-training is completed, adjusting and optimizing the parameters from top to bottom by using a BP algorithm.
5. The outpatient quantity prediction method based on the deep belief network as claimed in claim 4, wherein in the step S301, the network is trained by using a contrast divergence algorithm in a layer-by-layer training network using unlabeled data, specifically as follows:
s30101, sample data x is assigned to the visible layer according to a formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function；
S30102, extracting a sample h from the calculated probability distribution by adopting Gibbs sampling1～P(h1|v1)；
S30103, use1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1)；
Wherein, sigma also represents sigmoid activation function;
s30104, similarly, a Gibbs sample is taken from the probability distribution obtained by calculation to extract a sample v2～P(v2|h1)；
S30105, passing through v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
S30106, update the weight, W ← W + λ (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
6. The out-patient volume prediction method based on the deep belief network as claimed in claim 1, wherein the daily out-patient volume prediction is as follows:
(1) the input data is the outpatient service volume of each day in a certain selected time period, and the features among the outpatient service volume data are extracted through the constructed deep belief network, namely the input original data are converted into a group of feature vectors;
(2) after the characteristic vector is obtained, predicting the clinic diagnosis amount of the next day by utilizing a logistic regression layer and combining the characteristic vector, thereby obtaining a predicted value;
(3) and comparing the label data with the predicted value to obtain the difference between the label data and the predicted value, finely adjusting the parameters through back propagation, and obtaining the optimal group of parameters after limited iteration to realize the prediction of the clinic amount in a certain period of time in the future.
7. An outpatient quantity prediction system based on a deep belief network, the system comprising,
the depth confidence network construction unit is used for forming a depth confidence network based on the plurality of limited Boltzmann machines, the output trained by the limited Boltzmann machine of each layer is used as the input of the limited Boltzmann machine of the next layer to form the depth confidence network, and the whole depth confidence network is used as a data feature extraction layer for extracting the data features of the historical clinic amount;
a logistic regression layer establishing unit is used for adding a logistic regression layer on the upper layer of the deep confidence network to serve as a prediction layer, and carrying out supervised prediction on the clinic diagnosis amount data by combining the extracted data characteristics of the historical clinic diagnosis amount, so that the prediction precision is improved;
and the network training unit is used for training the neural network by using a greedy layer-by-layer algorithm.
8. The depth-confidence-network-based outpatient quantity prediction system of claim 7, wherein the bottom layer of the depth confidence network is a constrained Boltzmann machine, which consists of a visible layer and a hidden layer; the number of the neurons in the visible layer is set to be 6, the number of the neurons in the hidden layer is set to be 3, and the number of the layers in the hidden layer is set to be 3.
9. The depth-confidence-network-based outpatient prediction system of claim 7, wherein the logistic regression layer is a regression prediction layer superimposed on the depth confidence network, and the hospital outpatient is supervised predicted by using the data features extracted by the depth confidence network and the selected label data; the data features extracted by using the deep belief network are feature vectors converted from original input data through a neural network; the selection of tag data is related to the time interval.
10. The deep belief network-based outpatient prediction system of claim 7 or 8 or 9, wherein the network training unit comprises,
the data layer-by-layer training network module is used for training a network layer by using unlabeled data;
and the tuning module is used for tuning the parameters from top to bottom by using a BP algorithm after the pre-training is finished.
The data layer-by-layer training network module trains the network by using a contrast divergence algorithm, and the specific process is as follows:
①, sample data x is assigned to the visible layer according to the formula P (h)j|v)＝σ(bj+ΣiWi,jxi) Calculating the probability P (h) of each neuron in the hidden layer being activated1|v1) (ii) a Wherein σ represents a sigmoid activation function;
②, extracting a sample h from the calculated probability distribution by Gibbs sampling1～P(h1|v1)；
③, use h1Reconstructing the visible layer, i.e. back-pushing the visible layer by the hidden layer, according to the formula P (v)i|h)＝σ(ci+ΣjWi,jhj) Calculating the probability P (v) that each neuron in the visible layer is activated2|h1)；
Wherein, sigma also represents sigmoid activation function;
④, similarly, a sample v is extracted from the computed probability distribution by Gibbs sampling2～P(v2|h1)；
⑤, passing v2Calculating the activated probability of each neuron in the hidden layer again to obtain probability distribution P (h)2|v2)；
⑥, update weight, W ← W + lambda (P (h)1|v1)v1-P(h2|v2)v2)，b←b+λ(v1-v2)，c←c+λ(h1-h2)。
CN202010110803.8A 2020-02-24 2020-02-24 Outpatient quantity prediction method and system based on deep belief network Pending CN111276229A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010110803.8A CN111276229A (en) 2020-02-24 2020-02-24 Outpatient quantity prediction method and system based on deep belief network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010110803.8A CN111276229A (en) 2020-02-24 2020-02-24 Outpatient quantity prediction method and system based on deep belief network

Publications (1)

Publication Number Publication Date
CN111276229A true CN111276229A (en) 2020-06-12

Family

ID=71002263

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010110803.8A Pending CN111276229A (en) 2020-02-24 2020-02-24 Outpatient quantity prediction method and system based on deep belief network

Country Status (1)

Country Link
CN (1) CN111276229A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114202037A (en) * 2022-01-14 2022-03-18 上海蓬海涞讯数据技术有限公司 Generation method, prediction method, system and storage medium of hospital prediction model

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130339202A1 (en) * 2012-06-13 2013-12-19 Opera Solutions, Llc System and Method for Detecting Billing Errors Using Predictive Modeling
CN106326634A (en) * 2016-08-09 2017-01-11 浙江工业大学 Hospital outpatient quantity predicting method based on deep belief network
CN106407690A (en) * 2016-09-28 2017-02-15 厦门理工学院 Outpatient number prediction method and system based on automatic deep belief network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130339202A1 (en) * 2012-06-13 2013-12-19 Opera Solutions, Llc System and Method for Detecting Billing Errors Using Predictive Modeling
CN106326634A (en) * 2016-08-09 2017-01-11 浙江工业大学 Hospital outpatient quantity predicting method based on deep belief network
CN106407690A (en) * 2016-09-28 2017-02-15 厦门理工学院 Outpatient number prediction method and system based on automatic deep belief network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
杨旭华，钟楠祎: ""基于深度信念网络的医院门诊量预测"", 《计算机科学》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114202037A (en) * 2022-01-14 2022-03-18 上海蓬海涞讯数据技术有限公司 Generation method, prediction method, system and storage medium of hospital prediction model

Similar Documents

Publication Publication Date Title
CN113962364B (en) Multi-factor power load prediction method based on deep learning
Gu et al. A surrogate-assisted multi-objective particle swarm optimization of expensive constrained combinatorial optimization problems
Azadeh et al. Integration of artificial neural networks and genetic algorithm to predict electrical energy consumption
CN106022549A (en) Short term load predication method based on neural network and thinking evolutionary search
CN111127246A (en) Intelligent prediction method for transmission line engineering cost
CN101782976A (en) Automatic selection method for machine learning in cloud computing environment
CN109685240A (en) A kind of prediction of 95598 traffic work orders and unusual fluctuation method for early warning based on LSTM deep learning
CN106022521A (en) Hadoop framework-based short-term load prediction method for distributed BP neural network
Liu et al. Heating load forecasting for combined heat and power plants via strand-based LSTM
CN115689008A (en) CNN-BilSTM short-term photovoltaic power prediction method and system based on ensemble empirical mode decomposition
CN113554466A (en) Short-term power consumption prediction model construction method, prediction method and device
CN115034430A (en) Carbon emission prediction method, device, terminal and storage medium
CN113627070A (en) Short-term photovoltaic power prediction method
Colmenar et al. Automatic generation of models for energy demand estimation using Grammatical Evolution
CN113326919A (en) Traffic travel mode selection prediction method based on computational graph
CN115759415A (en) Power consumption demand prediction method based on LSTM-SVR
CN116340726A (en) Energy economy big data cleaning method, system, equipment and storage medium
CN112036598A (en) Charging pile use information prediction method based on multi-information coupling
Sen Time Series Prediction based on Improved Deep Learning
CN111276229A (en) Outpatient quantity prediction method and system based on deep belief network
CN114596726A (en) Parking position prediction method based on interpretable space-time attention mechanism
CN115994459B (en) Method, system and equipment for optimizing construction time sequence of mass power grid infrastructure projects
CN117114184A (en) Urban carbon emission influence factor feature extraction and medium-long-term prediction method and device
CN111697560A (en) Method and system for predicting load of power system based on LSTM
Yang et al. An improved deep echo state network inspired by tissue-like P system forecasting for non-stationary time series

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