Outpatient scheduling method and system based on deep learning

Abstract

The application discloses a kind of outpatient scheduling method and system based on deep learning. User outpatient information is obtained. Based on the user outpatient information, a first fusion feature vector is obtained. Through a first prediction network, based on the user outpatient information and the first fusion feature vector, the time variation law is detected. By detecting the positions of different users in different time points in a plurality of historical days, big data is used to find the queuing situation, thereby finding the first prediction network that can estimate the number of queues. The time point at which each user arrives at the detection room position is detected, and the number of queues and the number of transactions to be processed corresponding to the time point are found by the first prediction network. By judging multiple times, the shortest time point at which the detection report is obtained after queuing and processing and the user detection room position recommended to the user for sequential detection are jointly judged by using deep learning and the greedy algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a clinic scheduling method and system based on deep learning. BACKGROUND

[0002] In the traditional hospital outpatient management, after the patient goes to the clinic, he needs to be tested by blood routine, CT, etc. After obtaining the data of the body, the doctor is found again for diagnosis, which can more accurately judge the disease. However, due to the different number of people queuing in the blood collection room, CT room, etc., and the user needs to go to multiple detection clinics for multiple tests. It is difficult to flexibly adjust according to the actual patient demand and the use of medical resources. The hospital management personnel are difficult to master the real-time and accurate information of each detection clinic, which will lead to the congestion of some detection clinics and the almost empty of some detection clinics. Thus, it will lead to the long waiting time of the patient's second diagnosis and poor medical experience. The use efficiency of hospital resources is low, which cannot meet the growing demand for medical services. Therefore, there is an urgent need for a method and system that can intelligently plan the route to the detection clinic according to real-time data. SUMMARY

[0003] The purpose of the present application is to provide a clinic scheduling method and system based on deep learning, which solves the above problems existing in the prior art.

[0004] In a first aspect, the embodiments of the present application provide a clinic scheduling method based on deep learning, comprising:

[0005] Obtaining user clinic information; the user clinic information includes the current position of the user, the position of m user detection clinics; the position of the user detection clinic represents the position of the clinic where the user needs to be examined;

[0006] Based on the user clinic information, a first fusion feature vector is obtained; the first fusion feature vector represents the position and time features of the user to the m user detection clinic positions;

[0007] Through the first prediction network, based on the user clinic information and the first fusion feature vector, the time variation law is detected, and a recommended user clinic vector and a planned time length are obtained; the recommended user clinic vector represents the user detection clinic positions recommended to the user for detection in turn; the planned time length represents the shortest time to be waited after reaching the user detection clinic positions in the recommended user clinic vector in turn.

[0008] Optionally, the first fusion feature vector is obtained based on the user clinic information, comprising:

[0009] Based on the current position of the user and the positions of multiple user detection clinics, a first connected matrix, a second connected vector and a third connected vector are obtained;

[0010] Based on the first connectivity matrix, the second connectivity vector and the third connectivity vector, a feature is extracted to obtain a first fusion feature vector; the first fusion feature vector represents the features of the fused geographical position information, processing state and queuing state.

[0011] Optionally, the first connectivity matrix, the second connectivity vector and the third connectivity vector are obtained based on the current position of the user and the plurality of user detection clinic positions, and the method comprises the following steps:

[0012] The number of to-be-processed transactions, the processing speed, the number of queuing and the queuing speed are obtained; the number of to-be-processed transactions represents the number of to-be-processed transactions in one user detection clinic position; the processing speed represents the speed of processing transactions in one user detection clinic position; the number of queuing represents the number of people queuing in one user detection clinic position; and the queuing speed represents the speed of queuing in one user detection clinic position;

[0013] The distances between the current position of the user and the m user detection clinic positions are filled into the first connectivity matrix; the rows of the first connectivity matrix comprise the current position of the user and the m user detection clinic positions; and the columns of the first connectivity matrix comprise the current position of the user and the m user detection clinic positions;

[0014] The number of to-be-processed transactions of the m user detection clinic positions is divided by the corresponding processing speed to obtain the second connectivity vector;

[0015] The number of queuing of the m user detection clinic positions is divided by the corresponding queuing speed to obtain the third connectivity vector.

[0016] Optionally, the training method of the first prediction network comprises:

[0017] A plurality of training queuing data sets of days are obtained; the training queuing data set comprises training queuing numbers of a plurality of detection clinic positions at n queuing time points; the training queuing number represents the number of people queuing in one user detection clinic position at one time point of one detection clinic position; n represents that 24 hours are evenly divided into n parts; and the detection clinic position represents the positions of all clinics in a hospital;

[0018] Based on the detection clinic position and the training queuing number, a training fusion feature vector is obtained;

[0019] One detection clinic position, n / 12 queuing time points and corresponding training queuing numbers are used to construct a queuing detection vector; the number of elements of the queuing detection vector is 1+n / 12+n / 12;

[0020] input the queuing detection vector and the training fusion feature vector into a first prediction network to obtain a predicted queuing number; the predicted queuing number represents the number of people queuing at the same detection clinic position at a next queuing time point;

[0021] 1 detection clinic position corresponds to 1 first prediction network;

[0022] loss between the training queuing data at the next queuing time point and the predicted queuing number is calculated, and the first prediction network is trained through back propagation.

[0023] Optionally, the loss between the training queuing data at the next queuing time point and the predicted queuing number is calculated, and the first convolutional neural network, the first deep neural network, the second deep neural network and the third deep neural network are trained at the same time.

[0024] Optionally, the detection time variation law is detected based on the user outpatient information and the first fusion feature vector through the first prediction network to obtain a recommended user clinic vector and a planning time length, including:

[0025] a user speed, a current time point and a plurality of current queuing numbers are obtained; the user speed represents the speed of the user walking; the current time point represents a time point corresponding to the current position of the user; and the current queuing number represents the number of people queuing at a user detection clinic position at the current time point;

[0026] m first waiting time lengths, m first detection time points and m first predicted queuing numbers are obtained based on the user speed, the current time point, m current queuing numbers and m queuing speeds through the first prediction network;

[0027] m-1 second waiting time lengths, m-1 second detection time points and m-1 second predicted queuing numbers are obtained based on the user speed, the first detection time point, m-1 first predicted queuing numbers and m-1 queuing speeds through the first prediction network;

[0028] 1*2*3…(m-1)*m user routes and corresponding processing time lengths are obtained by traversing all user detection clinic positions; the user route represents a route passing through all user detection clinic positions; and the processing time length represents a sum of the first waiting time length, the second waiting time length,..., and the mth waiting time length on a user route;

[0029] the processing time length smaller than other processing time lengths is taken as a planning time length, and user detection clinic positions corresponding to a user route corresponding to the planning time length are taken as elements in a recommended user clinic vector in turn.

[0030] Optionally, the obtaining, by the first prediction network, the m first queuing time lengths, the m first detection time points and the m first predicted queuing quantities based on the user speed, the current time point, the m current queuing quantities and the m queuing speeds comprises:

[0031] dividing the distance from the current position of the user in the first connectivity matrix to the m user detection clinic positions by the user speed to obtain the m first time lengths; the first time length represents a time length for the current position of the user to reach one user clinic position;

[0032] obtaining the m first predicted queuing quantities based on the user detection clinic positions, the current time point and the corresponding current queuing quantities;

[0033] dividing the first predicted queuing quantity by the queuing speed to obtain the first queuing time length;

[0034] adding the current time point to the first time length and adding the first queuing time length to obtain the first detection time point; the m user clinic positions correspond to the m first detection time points;

[0035] dividing the number of to-be-processed transactions at the first detection point by the corresponding processing speed to obtain the first queuing time length; the m first detection time points correspond to the m first queuing time lengths.

[0036] Optionally, the obtaining, by the first prediction network, the m first queuing time lengths, the m first detection time points and the m first predicted queuing quantities based on the user speed, the current time point, the m current queuing quantities and the m queuing speeds comprises:

[0037] constructing a current queuing detection vector based on the user detection clinic positions, the current time point and the corresponding current queuing quantities;

[0038] obtaining the first predicted queuing quantity based on the current queuing detection vector and the trained first prediction network; the first predicted queuing quantity represents a queuing quantity at a time point after the first time length from the current time point;

[0039] the m user clinic positions correspond to the m first predicted queuing quantities.

[0040] Optionally, the extracting features based on the first connectivity matrix, the second connectivity vector and the third connectivity vector to obtain the first fusion feature vector comprises:

[0041] inputting the first connectivity matrix into the first convolutional neural network to extract features and obtain a first feature map;

[0042] inputting the first feature map into the first deep neural network to extract features and obtain a first feature vector;

[0043] Input the first connected matrix into the second deep neural network, extract features, and obtain a second feature vector;

[0044] Input the first connected matrix into the third deep neural network, extract features, and obtain a third feature vector;

[0045] The first feature vector, the second feature vector and the third feature vector have the same number of elements;

[0046] The first feature vector, the second feature vector and the third feature vector are averaged to obtain a first fusion feature vector.

[0047] In a second aspect, an embodiment of the present application provides a deep learning-based outpatient scheduling system, comprising:

[0048] An acquisition module is configured to acquire user outpatient information; the user outpatient information comprises a user current position and m user detection room positions; the user detection room positions represent positions of detection rooms in which the user needs to be detected;

[0049] A feature module is configured to obtain a first fusion feature vector based on the user outpatient information; the first fusion feature vector represents features of positions and times of the user to the m user detection room positions;

[0050] A detection module is configured to detect a time variation law based on the user outpatient information and the first fusion feature vector through a first prediction network, and obtain a recommended user room vector and a planned time length; the recommended user room vector represents user detection room positions recommended to the user for detection in sequence; and the planned time length represents a shortest time that needs to be waited after the user detection room positions in the recommended user room vector are reached in sequence.

[0051] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0052] The embodiment of the present application also provides a deep learning-based outpatient scheduling method and system.

[0053] In the present application, the time for the user to be detected is estimated according to the number of to-be-handled transactions, the number of queues, the processing speed and the queuing speed of different user detection room positions at different time points. Through different user detection room positions at different time points in a plurality of days in the past, big data is used to find the queuing situation, so that a first prediction network capable of estimating the number of queues is found. The time point at which each user detection room position is reached is detected, and the number of queues and the number of to-be-handled transactions corresponding to the time point are found. Through multiple judgments, the shortest time point at which a detection report is obtained after queuing and processing is jointly judged by using deep learning and a greedy algorithm. Thus, the technical effect of saving the waiting time for detection can be achieved. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a deep learning-based outpatient scheduling method flowchart provided by an embodiment of the application. DETAILED DESCRIPTION

[0055] The application will be described in detail below with reference to the accompanying drawings.

[0056] Embodiment 1

[0057] As Figure 1 shown, the application provides a deep learning-based outpatient scheduling method, which comprises:

[0058] S101: Obtain user outpatient information; the user outpatient information comprises a user current position and m user detection room positions; the user detection room position represents the position of a room where the user needs to be examined.

[0059] In this embodiment, the user detection room position comprises a clinical laboratory position, an imaging department position, and a functional examination department position.

[0060] The clinical laboratory is used for blood routine examination, urine routine examination, and stool routine examination, etc. The imaging department is used for x-ray examination, CT, magnetic resonance imaging, and ultrasound, etc. The functional examination department is used for electrocardiogram and electromyogram, etc.

[0061] S102: Obtain a first fusion feature vector based on the user outpatient information; the first fusion feature vector represents the position and time features of the user to the m user detection room positions.

[0062] S103: Detect the time variation law based on the user outpatient information and the first fusion feature vector through a first prediction network, to obtain a recommended user room vector and a planning time length; the recommended user room vector represents the user detection room positions recommended to the user for detection in sequence; the planning time length represents the shortest time that needs to be waited after sequentially reaching the user detection room positions in the recommended user room vector.

[0063] Optionally, obtaining the first fusion feature vector based on the user outpatient information comprises:

[0064] Obtaining a first connectivity matrix, a second connectivity vector, and a third connectivity vector based on the user current position and the plurality of user detection room positions.

[0065] Extracting features based on the first connectivity matrix, the second connectivity vector, and the third connectivity vector to obtain the first fusion feature vector; the first fusion feature vector represents the features of fused geographic position information, processing state, and queuing state.

[0066] Optionally, the first connectivity matrix, the second connectivity vector and the third connectivity vector are obtained based on the current location of the user and the plurality of user detection clinic locations, comprising:

[0067] The number of to-be-processed transactions, the processing speed, the number of people in the queue and the queue speed are obtained; the number of to-be-processed transactions represents the number of to-be-processed transactions in one user detection clinic location; the processing speed represents the speed of processing transactions in one user detection clinic location; the number of people in the queue represents the number of people in the queue in one user detection clinic location; and the queue speed represents the speed of people in the queue in one user detection clinic location.

[0068] The number of people in the queue can be obtained by taking a number. The queue speed represents the speed of users being called for a number.

[0069] The to-be-processed transactions in the user detection clinic location corresponding to the blood collection room are blood tests. The number of to-be-processed transactions represents the number of samples that have not been subjected to blood tests. The processing speed represents the speed of blood tests.

[0070] The number of people in the queue changes with the change of time points.

[0071] The distances between the current location of the user and the m user detection clinic locations are filled into the first connectivity matrix; the rows of the first connectivity matrix include the current location of the user and the m user detection clinic locations; and the columns of the first connectivity matrix include the current location of the user and the m user detection clinic locations.

[0072] The partial structure of the first connectivity matrix is shown in Table 1.

[0073]

[0074] The values in the first connectivity matrix are in meters.

[0075] The values in the first connectivity matrix represent the distances between the current location of the user and the m user detection clinic locations and the distances between the m user detection clinic locations.

[0076] The distances are not straight-line distances, but the lengths of corresponding routes.

[0077] The number of to-be-processed transactions in the m user detection clinic locations is divided by the corresponding processing speed to obtain the second connectivity vector.

[0078] The values in the second connectivity vector are in minutes.

[0079] The number of people in the queue in the m user detection clinic locations is divided by the corresponding queue speed to obtain the third connectivity vector.

[0080] wherein the value in the second communication vector is in units of meters / minute.

[0081] Optionally, the training method of the first prediction network comprises:

[0082] obtaining a plurality of sets of training queuing data for a plurality of days; each set of training queuing data comprising a plurality of training queuing quantities for a plurality of detection clinic positions at n queuing time points; each training queuing quantity representing the number of people queuing for a detection clinic position at a time point; n representing the division of 24 hours into n parts; and the detection clinic position representing all clinic positions in a hospital.

[0083] wherein n is a positive integer, and in this embodiment, n is 72; and the 24 hours are divided into 72 parts, each part representing 20 minutes.

[0084] wherein the user detection clinic position is a clinic position that a user needs to go to for detection, and the detection clinic position is all clinic positions in a hospital, and the two have different meanings.

[0085] obtaining a training fusion feature vector based on the detection clinic position and the training queuing quantity.

[0086] wherein the method of obtaining the training fusion feature vector is the same as the method of obtaining the first fusion feature vector.

[0087] constructing a queuing detection vector for a detection clinic position, n / 12 queuing time points, and corresponding training queuing quantities; the queuing detection vector having 1+n / 12+n / 12 elements.

[0088] wherein the first element of the queuing detection vector is the detection clinic position, the second element is the queuing time point, the third element is the training queuing quantity corresponding to the queuing time point corresponding to the second element, the fourth element is the next queuing time point of the queuing time point corresponding to the second element, the fifth element is the training queuing quantity corresponding to the queuing time point corresponding to the fourth element, …, the nth / 12+n / 12 element is the next queuing time point corresponding to the queuing time point corresponding to the nth / 12+n / 12-2 element, and the 1+n / 12+n / 12 element is the training queuing quantity corresponding to the queuing time point corresponding to the nth / 12+n / 12 element.

[0089] wherein the queuing time point and the corresponding next queuing time point are adjacent time points.

[0090] wherein n is greater than 12.

[0091] Input the queuing detection vector and the training fusion feature vector into a first prediction network to obtain a predicted queuing number; the predicted queuing number represents the number of people queuing at the same detection clinic position at a next queuing time point.

[0092] In general, the number of people queuing in the hospital is regular, such as the number of people queuing on weekends is higher than that on weekdays, and the number of people queuing in the afternoon is higher than that in the morning. And the number of people queuing at each detection clinic position is different, such as the number of people queuing at the detection clinic position of blood routine is more than that at the detection clinic position of CT. Therefore, the detection clinic position, the queuing time point and the corresponding training queuing number are used to find the regularity together, so as to find the queuing number of the same detection clinic position at the next queuing time point.

[0093] 1 detection clinic position corresponds to 1 first prediction network;

[0094] The loss is calculated by the training queuing data at the next queuing time point and the predicted queuing number, and the first prediction network is trained by back propagation.

[0095] In this embodiment, the loss is calculated by using a cross-entropy loss function.

[0096] Optionally, the loss is calculated by the training queuing data at the next queuing time point and the predicted queuing number, and the first convolutional neural network, the first deep neural network, the second deep neural network and the third deep neural network are trained at the same time.

[0097] Optionally, the time variation rule is detected by the first prediction network based on the user outpatient information and the first fusion feature vector to obtain a recommended user clinic vector and a planned time length, including:

[0098] Obtain the user speed, the current time point and the plurality of current queuing numbers; the user speed represents the speed of the user walking; the current time point represents the time point corresponding to the current position of the user; and the current queuing number represents the number of people queuing at a user detection clinic position at the current time point.

[0099] The first prediction network is used to obtain m first waiting time lengths, m first detection time points and m first predicted queuing numbers based on the user speed, the current time point, m current queuing numbers and m queuing speeds.

[0100] Each first waiting time length corresponds to a first detection time point, a first predicted queuing number and a user detection clinic position.

[0101] Obtaining m-1 second waiting time lengths, m-1 second detection time points and m-1 second predicted queuing quantities based on the user speed, the first detection time point, m-1 first predicted queuing quantities and m-1 queuing speeds through the first prediction network.

[0102] Wherein, the m-1 first predicted queuing quantities and the m-1 queuing speeds are the first predicted queuing quantities and the queuing speeds of the m-1 user detection clinic positions except the user detection clinic position corresponding to the first detection time point.

[0103] Wherein, the 1 second waiting time length corresponds to the 1 second detection time point, the 1 second predicted queuing quantity corresponds to the 1 user detection clinic position.

[0104] Wherein, the method for obtaining the m-1 second waiting time lengths, the m-1 second detection time points and the m-1 second predicted queuing quantities is the same as the method for obtaining the m first waiting time lengths, the m first detection time points and the m first predicted queuing quantities.

[0105] Obtaining 1*2*3…(m-1)*m user routes and corresponding processing time lengths by traversing all user detection clinic positions; the user route represents a route passing through all user detection clinic positions; the processing time length represents the sum of the first waiting time length, the second waiting time length,..., the m waiting time length on 1 user route.

[0106] Taking the processing time length smaller than other processing time lengths as a planning time length, and taking the user detection clinic positions corresponding to the user route corresponding to the planning time length as elements in the recommended user clinic vector in turn.

[0107] Optionally, the method for obtaining m first waiting time lengths, m first detection time points and m first predicted queuing quantities based on the user speed, the current time point, m current queuing quantities and m queuing speeds through the first prediction network comprises:

[0108] Dividing the distance from the user current position to the m user detection clinic positions in the first connected matrix by the user speed to obtain m first time lengths; the first time length represents the time length for the user current position to reach 1 user clinic position.

[0109] Obtaining m first predicted arrangement quantities based on the user detection clinic position, the current time point and the corresponding current queuing quantity.

[0110] Dividing the first predicted queuing quantity by the queuing speed to obtain a first arrangement time length;

[0111] Add the current time point, the first time length and the first arrangement time length to obtain a first detection time point; m user detection room positions correspond to obtain m first detection time points;

[0112] Divide the number of to-be-processed transactions at the first detection point by the corresponding processing speed to obtain a first waiting time length; m first detection time points correspond to obtain m first waiting time lengths.

[0113] Optionally, the m first predicted arrangement quantities are obtained based on the user detection room positions, the current time point and the corresponding current queue number, and include:

[0114] The user detection room positions, the current time point and the corresponding current queue number are constructed into a current queue detection vector.

[0115] In the embodiment, if n=72, then 72 / 24=3, and therefore the number of elements of the current queue detection vector in the embodiment is 1+3*2=7 elements. Because the information before the current time point is unknown, the current time point is 8:00 am, and therefore the constructed current queue detection vector is [02, 0, 0, 0, 0, 01, 24], indicating that the user detection room position with the label 02 has queued 24 people at 8:00 am with the label 01.

[0116] The first predicted queue number is obtained by using the trained first prediction network according to the current queue detection vector; the first predicted queue number represents the number of people queuing at a time point after the first time length from the current time point.

[0117] The next predicted number is obtained by inputting the current queue detection vector into the trained first prediction network; the next predicted number represents the number of people queuing at the next queuing time point of the current time point. If the current time point is 8:00 am, because n is 72, the next queuing time point is 8:20. Therefore, the number of people queuing at 8:20 is found by one prediction. The time point after adding the first time length to the current time point is taken as the first time point. If the detected next predicted number is 42, a next queue detection vector is constructed by using the user detection room positions, the first time point and the corresponding next predicted number. The next queue detection vector is [02, 0, 0, 01, 24, 02, 46]. The next predicted number is obtained by inputting the next queue detection vector into the trained first prediction network; the next predicted number represents the number of people queuing at 8:40.

[0118] If the first time length is 5 minutes, the number of people queuing at the closest 8:00 is taken as the first predicted queuing number. If the first time length is 10 minutes, the number of people queuing at the closest 8:00 is taken as the first predicted queuing number. If the first time length is 15 minutes, the number of people queuing at the closest 8:20 is taken as the first predicted queuing number. If the first time length is 35 minutes, the number of people queuing at the closest 8:40 is taken as the first predicted queuing number. The number of people queuing at a time point closest to the first time length is found.

[0119] The first prediction network is a deep neural network (DNN) containing 5 layers of neuron layers.

[0120] The m first predicted queuing numbers are obtained corresponding to the m user clinic positions.

[0121] Optionally, the first fusion feature vector is obtained by extracting features based on the first connectivity matrix, the second connectivity vector and the third connectivity vector, and the method comprises:

[0122] The first connectivity matrix is input into a first convolutional neural network to extract features and obtain a first feature map.

[0123] In this embodiment, the first convolutional neural network is a convolutional neural network (CNN) composed of 6 layers of 2*2 convolution kernels.

[0124] The first feature map is input into a first deep neural network to extract features and obtain a first feature vector.

[0125] In this embodiment, the first deep neural network is a deep neural network (DNN) containing 3 layers of neuron layers.

[0126] The first connectivity matrix is input into a second deep neural network to extract features and obtain a second feature vector.

[0127] In this embodiment, the second deep neural network is a deep neural network (DNN) containing 5 layers of neuron layers.

[0128] The first connectivity matrix is input into a third deep neural network to extract features and obtain a third feature vector.

[0129] In this embodiment, the third deep neural network is a deep neural network (DNN) containing 6 layers of neuron layers.

[0130] The first feature vector, the second feature vector and the third feature vector have the same number of elements.

[0131] The first deep neural network, the second deep neural network and the third deep neural network have the same number of neurons in the output layer.

[0132] The first feature vector, the second feature vector and the third feature vector are averaged to obtain a first fusion feature vector.

[0133] Embodiment 2

[0134] Based on the above-mentioned deep learning-based outpatient scheduling method, the embodiment of the present application further provides a deep learning-based outpatient scheduling system, the system comprising:

[0135] An acquisition module is configured to acquire user outpatient information, wherein the user outpatient information comprises a current position of a user and m user detection room positions, and the user detection room positions represent positions of detection rooms that the user needs to detect.

[0136] A feature module is configured to obtain a first fusion feature vector based on the user outpatient information, wherein the first fusion feature vector represents features of positions and times of the user to the m user detection room positions.

[0137] A detection module is configured to detect a time variation rule based on the user outpatient information and the first fusion feature vector through a first prediction network, and obtain a recommended user room vector and a planned time length, wherein the recommended user room vector represents user detection room positions recommended to the user in sequence, and the planned time length represents a shortest time that needs to be waited after the user reaches the user detection room positions in the recommended user room vector in sequence.

[0138] The algorithms and displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general purpose systems can be used with programs in accordance with the teachings herein, or it can prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will be apparent from the description above. In addition, the present application is not intended to be limited to a particular programming language. It will be appreciated that a variety of programming languages can be used to implement the teachings of the application as described herein, and any references below to specific languages are provided for disclosure of enablement of the best mode of the application.

[0139] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order to avoid obscuring the understanding of this description.

[0140] The various component embodiments of the present application can be implemented in hardware, or as software modules running in one or more processors, or in combinations thereof. As will be appreciated by one skilled in the art, a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functionality of some or all of the components in an apparatus according to an embodiment of the present application. The present application can also be implemented as an apparatus or device program (e.g., computer program and computer program product) for performing part or all of the methods described herein. A program that implements the present application can be stored on a computer-readable medium, or can be transmitted over one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier medium, or in any other form.

Claims

1. A deep learning-based outpatient scheduling method, characterized in that, include: Obtain user outpatient information; The user outpatient information includes the user's current location and the locations of m user testing clinics; The user's examination room location refers to the location of the examination room where the user needs to undergo the examination; Based on the user's outpatient information, a first fusion feature vector is obtained; the first fusion feature vector represents the location and time characteristics of the user's journey to the m user testing clinic locations; Using the first prediction network, based on the user's outpatient information and the first fused feature vector, the time change pattern is detected to obtain the recommended user clinic vector and the planned time length; the recommended user clinic vector represents the user clinic location recommended to the user for sequential detection; the planned time length represents the shortest waiting time required after sequentially arriving at the user clinic location in the recommended user clinic vector. The first fusion feature vector obtained based on the user's outpatient information includes: Based on the user's current location and the locations of multiple user detection clinics, a first connectivity matrix, a second connectivity vector, and a third connectivity vector are obtained. Based on the first connected matrix, the second connected vector, and the third connected vector, features are extracted to obtain the first fused feature vector; the first fused feature vector represents the features of fused geographic location information, processing status, and queuing status. The training method for the first prediction network includes: Obtain a training queuing data set for multiple days; the training queuing data set includes the number of training queues for multiple testing clinic locations at n queuing time points; the number of training queues represents the number of people queuing for one user testing clinic location at one time point; n represents dividing 24 hours into n equal parts; the testing clinic location represents the location of all clinics in the hospital; Based on the location of the detection clinic and the number of training queues, a training fusion feature vector is obtained; Construct a queue detection vector by taking one testing clinic location, n / 12 queuing time points, and the corresponding training queue number; the number of elements in the queue detection vector is 1 + n / 12 + n / 12. The queue detection vector and the trained fusion feature vector are input into the first prediction network to obtain the predicted queue number; the predicted queue number represents the number of people queuing at the next queuing time point for the same testing clinic location; One testing clinic location corresponds to one first prediction network; The loss is calculated by combining the training queuing data and the predicted queuing number at the next queuing time point, and then backpropagated to train the first prediction network. The step of using a first prediction network, based on the user's outpatient information and a first fused feature vector, to detect time variation patterns and obtain the recommended user's clinic vector and planned time length includes: The system obtains user speed, current time point, and multiple current queue numbers; the user speed represents the user's walking speed; the current time point represents the time point corresponding to the user's current location; and the current queue number represents the number of people queuing at the current time point for one user's examination room location. Through the first prediction network, based on the user speed, the current time point, m current queue counts and m queue speeds, m first waiting time lengths, m first detection time points and m first predicted queue counts are obtained; Based on the user speed, the first detection time point, m-1 first predicted queue lengths and m-1 queue speeds, the first prediction network obtains m-1 second waiting time lengths, m-1 second detection time points, and m-1 second predicted queue lengths. Traverse all user testing clinic locations to obtain 1*2*3…(m-1)*m user routes and their corresponding processing time lengths; the user route represents the route that passes through all user testing clinic locations; the processing time length represents the sum of the first waiting time length, the second waiting time length, ..., the m-th waiting time length on a user route; The processing time length that is shorter than other processing time lengths is taken as the planning time length, and the user detection clinic location corresponding to the user route corresponding to the planning time length is sequentially used as the element in the recommended user clinic vector.

2. The outpatient scheduling method based on deep learning according to claim 1, characterized in that, The process of obtaining a first connected matrix, a second connected vector, and a third connected vector based on the user's current location and the locations of multiple user detection clinics includes: The system obtains the number of pending transactions, processing speed, queue size, and queue speed; the number of pending transactions represents the number of pending transactions in a user's clinic location; the processing speed represents the processing speed of transactions in a user's clinic location; the queue size represents the number of people queuing in a user's clinic location; and the queue speed represents the queuing speed in a user's clinic location. Fill the distances between the user's current location and the locations of the m user testing rooms into the first connected matrix; the rows of the first connected matrix contain the user's current location and the locations of the m user testing rooms; the columns of the first connected matrix contain the user's current location and the locations of the m user testing rooms. Divide the number of pending transactions for detecting clinic locations for m users by the corresponding processing speed to obtain the second connected vector; Divide the queue length of m users for detecting clinic locations by the corresponding queue speed to obtain the third connected vector.

3. The outpatient scheduling method based on deep learning according to claim 1, characterized in that, The loss is calculated by combining the training queuing data and the predicted queuing number at the next queuing time point, and the first convolutional neural network, the first deep neural network, the second deep neural network, and the third deep neural network are trained simultaneously.

4. The outpatient scheduling method based on deep learning according to claim 1, characterized in that, The step of obtaining m first waiting time lengths, m first detection time points, and m first predicted queue numbers through the first prediction network, based on the user speed, the current time point, m current queue numbers, and m queue speeds, includes: Divide the distance from the user's current position to the m user detection room positions in the first connected matrix by the user's speed to obtain m first time lengths; the first time length represents the time length for the user's current position to reach one user detection room position; Based on the user's detected clinic location, current time, and corresponding current queue number, m first predicted permutations are obtained; Divide the first predicted queue size by the queue speed to obtain the first queueing time length; Add the current time point to the first time length and the first arrangement time length to obtain the first detection time point; m user clinic locations correspond to m first detection time points; The number of pending transactions at the first detection point is divided by the corresponding processing speed to obtain the first waiting time length; m first detection time points correspond to m first waiting time lengths.

5. The outpatient scheduling method based on deep learning according to claim 4, characterized in that, The method, based on the user's detected clinic location, current time, and corresponding current queue number, yields m first predicted permutations, including: Construct the current queue detection vector by detecting the user's clinic location, current time, and corresponding current queue number; Based on the current queue detection vector, the first predicted queue number is obtained through the trained first prediction network; the first predicted queue number represents the number of queues that have been queued after a first time length from the current time point. For each of the m user clinic locations, m first predicted permutations are obtained.

6. The outpatient scheduling method based on deep learning according to claim 1, characterized in that, The first fused feature vector is obtained by extracting features based on the first connected matrix, the second connected vector, and the third connected vector, including: The first connected matrix is ​​input into the first convolutional neural network to extract features and obtain the first feature map; The first feature map is input into the first deep neural network to extract features and obtain the first feature vector; The first connected matrix is ​​input into the second deep neural network to extract features and obtain the second feature vector; The first connected matrix is ​​input into the third deep neural network to extract features and obtain the third feature vector; The first feature vector, the second feature vector, and the third feature vector have the same number of elements; The first fused feature vector is obtained by averaging the corresponding indices of the first, second, and third feature vectors.

7. A deep learning-based outpatient scheduling system, characterized in that, include: The acquisition module is used to acquire user outpatient information; The user outpatient information includes the user's current location and the locations of m user testing clinics; The user's examination room location refers to the location of the examination room where the user needs to undergo the examination; The feature module is used to obtain a first fused feature vector based on the user's outpatient information; the first fused feature vector represents the location and time characteristics of the user's journey to the m user testing clinic locations; The detection module is used to detect time change patterns through a first prediction network, based on the user's outpatient information and a first fusion feature vector, to obtain a recommended user clinic vector and a planned time length; the recommended user clinic vector represents the location of the user clinic recommended to the user for sequential detection; the planned time length represents the shortest waiting time required after sequentially reaching the user clinic location in the recommended user clinic vector. The first fusion feature vector obtained based on the user's outpatient information includes: Based on the user's current location and the locations of multiple user detection clinics, a first connectivity matrix, a second connectivity vector, and a third connectivity vector are obtained. Based on the first connected matrix, the second connected vector, and the third connected vector, features are extracted to obtain the first fused feature vector; the first fused feature vector represents the features of fused geographic location information, processing status, and queuing status. The training method for the first prediction network includes: Obtain a training queuing data set for multiple days; the training queuing data set includes the number of training queues for multiple testing clinic locations at n queuing time points; the number of training queues represents the number of people queuing for one user testing clinic location at one time point; n represents dividing 24 hours into n equal parts; the testing clinic location represents the location of all clinics in the hospital; Based on the location of the detection clinic and the number of training queues, a training fusion feature vector is obtained; Construct a queue detection vector by taking one testing clinic location, n / 12 queuing time points, and the corresponding training queue number; the number of elements in the queue detection vector is 1 + n / 12 + n / 12. The queue detection vector and the trained fusion feature vector are input into the first prediction network to obtain the predicted queue number; the predicted queue number represents the number of people queuing at the next queuing time point for the same testing clinic location; One testing clinic location corresponds to one first prediction network; The loss is calculated by combining the training queuing data and the predicted queuing number at the next queuing time point, and then backpropagated to train the first prediction network. The step of using a first prediction network, based on the user's outpatient information and a first fused feature vector, to detect time variation patterns and obtain the recommended user's clinic vector and planned time length includes: The system obtains user speed, current time point, and multiple current queue numbers; the user speed represents the user's walking speed; the current time point represents the time point corresponding to the user's current location; and the current queue number represents the number of people queuing at the current time point for one user's examination room location. Through the first prediction network, based on the user speed, the current time point, m current queue counts and m queue speeds, m first waiting time lengths, m first detection time points and m first predicted queue counts are obtained; Based on the user speed, the first detection time point, m-1 first predicted queue lengths and m-1 queue speeds, the first prediction network obtains m-1 second waiting time lengths, m-1 second detection time points, and m-1 second predicted queue lengths. Traverse all user testing clinic locations to obtain 1*2*3…(m-1)*m user routes and their corresponding processing time lengths; the user route represents the route that passes through all user testing clinic locations; the processing time length represents the sum of the first waiting time length, the second waiting time length, ..., the m-th waiting time length on a user route; The processing time length that is shorter than other processing time lengths is taken as the planning time length, and the user detection clinic location corresponding to the user route corresponding to the planning time length is sequentially used as the element in the recommended user clinic vector.

Images