CN-122000007-A - Dynamic optimization method, device, equipment and medium for medical treatment visit resource reservation

CN122000007ACN 122000007 ACN122000007 ACN 122000007ACN-122000007-A

Abstract

The invention discloses a dynamic optimization method, device, equipment and medium for medical treatment visit resource reservation, which comprise the steps of acquiring and preprocessing multidimensional panoramic data, constructing a bimodal patient model with static clinical portraits and dynamic state evolution, combining a dynamic weight clinical index evaluation and clinical priority dynamic quantification method to obtain clinical characteristics of a patient, obtaining a detection duration prediction result, constructing a detection operation constraint matrix to obtain a detection operation constraint rule, constructing a MADDPG multi-agent resource scheduling model under CTDE framework, obtaining a resource scheduling joint optimization strategy through multi-agent joint training, generating a reservation scheme based on the resource scheduling joint optimization strategy, and performing rolling update on the reservation scheme based on a prospective rolling rescheduling mechanism. The invention belongs to the field of diagnostic strategy optimization. The invention can provide more reasonable appointment strategy for the doctor.

Inventors

LIU XUEMEI
YUE XIAOBO
ZHANG PING

Assignees

四川互慧软件有限公司

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A dynamic optimization method for medical treatment resource reservation, which is characterized by comprising the following steps: Acquiring and preprocessing multidimensional panoramic data, wherein preprocessing comprises data cleaning, standardization and time-space feature joint coding; Based on the preprocessed multidimensional panoramic data, constructing a bimodal patient model of static clinical portraits and dynamic state evolution, and combining a dynamic weight clinical index evaluation and clinical priority dynamic quantification method to obtain clinical characteristics of patients; Constructing a multi-dimensional coupling feature set based on clinical features of a patient, inputting the multi-dimensional coupling feature set into an examination duration prediction model with an online self-correction mechanism and based on a transducer attention mechanism to obtain an examination duration prediction result, and constructing an examination operation constraint matrix to obtain an examination operation constraint rule; Based on the checking duration prediction result and the checking operation constraint rule, constructing a MADDPG multi-agent resource scheduling model under CTDE framework, and obtaining a resource scheduling joint optimization strategy through multi-agent joint training; and generating a reservation scheme based on the resource scheduling combined optimization strategy, and carrying out rolling update on the reservation scheme based on a prospective rolling rescheduling mechanism.
2. The dynamic optimization method for medical treatment resource reservation according to claim 1, wherein constructing a bimodal patient model of static clinical portraits and dynamic state evolution based on preprocessed multidimensional panoramic data, and combining a dynamic weighted clinical index evaluation and a clinical priority dynamic quantification method to obtain clinical characteristics of patients, comprises: Constructing a static clinical portrayal model comprising: Wherein, the As a result of the static feature vector, For the feature vector of the basic information of the patient, Is the characteristic vector of the past diseases, allergies, complications and medical history of patients, Feature vectors are constrained for patient clinical paths and examination items, Reserving a performance rate and checking fit characteristic vector for patient history; constructing a dynamic state evolution model, comprising: based on real-time physiological index data, constructing time sequence dynamic feature vector ; Based on time sequence dynamic characteristic vector Outputting a dynamic state evolution model, comprising: Wherein, the Is that The hidden layer state of the LSTM network at the moment, In order to be the hidden layer state at the previous moment, For the weight matrix of the LSTM network, Is that The dynamic state vector of the patient at the moment, The function is activated for Sigmoid, Is a weight matrix of the full connection layer, Bias terms for the full connection layer; A dynamic weighted clinical index assessment system, wherein the dynamic weighted clinical index assessment system is associated with a treatment stage and a disease type; under the dynamic weight clinical index evaluation system, carrying out clinical priority dynamic quantification, which comprises the following steps: Wherein, the Is that The clinical priority score for the time of day, 、 And (3) with Weights provided for a dynamic weighted clinical index assessment system, For the severity score of the illness state, In order to check the timeliness score, Constraining scores for clinical paths; Wherein, the Wherein, the For the longest period of the inspection window, For the remaining checkable time that the current distance window period expires, Is the disease aging coefficient.
3. The method for dynamically optimizing medical treatment resource reservation according to claim 1, wherein constructing a multi-dimensional coupling feature set based on clinical features of a patient, inputting the multi-dimensional coupling feature set into a examination duration prediction model based on a transducer attention mechanism with an online self-correction mechanism to obtain an examination duration prediction result, and constructing an examination operation constraint matrix to obtain an examination operation constraint rule, comprises: Based on a multi-headed self-attention mechanism, determining the association weights between different features in a multi-dimensional coupling feature set constructed by clinical features of a patient comprises: Wherein, the In order to query the matrix, In the form of a matrix of keys, In the form of a matrix of values, For the purpose of the transposition, Is a feature dimension; outputting the mean value and standard deviation of the inspection duration prediction result through the full connection layer, wherein the method comprises the following steps: Wherein, the For the multi-dimensional coupling of feature vectors, In the form of a transducer encoder, In order to provide the encoder weight matrix, To check the mean value of the duration predictions, To check the standard deviation of the duration prediction result, As the weight of the mean value output layer, For the bias of the mean value output layer, The weights of the layers are output for the variance, Bias for variance output layer; Based on the mean and standard deviation, the confidence interval is determined as: Constructing a loss function, comprising: Wherein, the As a function of the loss, For the number of samples to be taken, Is the first The actual length of the examination of the individual samples, Is the first The examination duration of each sample predicts the outcome, For the regularization coefficient of L2, The model weight matrix is adopted; constructing inspection operation constraint matrix, including setting minimum operation interval for different inspection combinations to obtain Constraint matrix of dimensions 。
4. The dynamic optimization method for medical treatment resource reservation according to claim 1, wherein a MADDPG multi-agent resource scheduling model under CTDE framework is constructed based on the prediction result of the examination duration and the constraint rule of the examination operation, and a resource scheduling joint optimization strategy is obtained through multi-agent joint training, comprising: construction of patient agent clusters Comprising: Wherein, the Is the first A patient agent; constructing resource agent clusters Comprising: Wherein, the Is the first A resource agent; Defining elements of a Markov decision, including a global state space, a local observation space, an action space, a reward function, and a state transition probability; Designing a reward function that constructs a patient agent cluster, comprising: Wherein, the The prize value for the patient agent, 、 And Are all the weights of the materials, Is the first The expected wait time period for each sample, Is the first The maximum acceptable waiting period for each sample, Is the first The assigned reserved time slots of the individual samples, Is the first The ideal reservation time for each sample, Is the first An acceptable reservation time window for each sample, Is the first Clinical priority scores for the individual samples are determined, Is the first Historical appointment compliance scores for the individual samples; Designing a reward function for constructing a resource agent cluster, including: Wherein, the To be the rewarding value of the resource agent, Is resource intelligent agent At the position of The utilization rate of the time period is set, For an optimal utilization of the resources, Is resource intelligent agent At the position of The cumulative load of the time period, Is resource intelligent agent Is used for the load balancing of the (c), For the maximum rated load of the resource, To utilize resource intelligent agent Is of medical staff The real-time fatigue of the time period, 、 And Are all weights; Global optimization based on a reward function, comprising: Wherein, the As a result of the global weight being given, In order to optimize the result of the search, For the number of samples to be taken, Is the number of resource agents; performing policy gradient updates, including: Wherein, the First, the Policy network parameters for the individual samples, In order to be desirable for the jackpot to accumulate, For the experience playback pool, Is the first A local observation of the individual samples is made, In the case of a global state, For a centralized action cost function based on global information, Is the first A policy network of the individual samples, Is the first The space of action of the individual samples, In the form of a global state space, Is desirable.
5. The method of claim 1, wherein generating a reservation plan based on the resource scheduling joint optimization strategy and rolling updating the reservation plan based on a look-ahead rolling rescheduling mechanism comprises: the resource scheduling is combined with the optimization strategy and the inspection operation constraint rule to generate a diagnosis and treatment node sequence, and a reservation scheme is generated through priority sequencing; performing rolling optimization with the optimization objective of minimizing the increase of the predicted waiting time length, minimizing the reserved adjustment quantity and maximizing the resource utilization rate, including: Wherein, the In order to scroll through the optimization function, 、 And Are all the weights of the materials, In order to anticipate the increase in the waiting time period, As a global fluctuation value of the resource utilization, The quantity is adjusted for the reservation.
6. The dynamic optimization method for medical treatment resource reservation as claimed in claim 1, further comprising: constructing a three-dimensional effect evaluation system, which comprises the following steps: Wherein, the For the global evaluation of the score(s), 、 And Are all the weights of the materials, For the clinical dimension to normalize the score, In order to normalize the score for the operational dimension, Normalizing the score for the patient dimension; To maximize the For the target, adopting a Bayesian optimization algorithm to carry out self-adaptive optimization on global parameters, wherein the self-adaptive optimization comprises the following steps: Wherein, the Is a global policy network parameter.
7. The method for dynamic optimization of medical care resource reservation as recited in claim 1, wherein preprocessing the multi-dimensional panoramic data comprises: Classifying the multi-dimensional panoramic data, wherein the data types comprise continuous characteristic data, classified characteristic data and ordered characteristic data; Filling missing values and removing abnormal values of the multi-dimensional panoramic data, performing single-heat coding on the split type characteristic data, and performing label coding on the ordered type characteristic data; carrying out dimensionless treatment on the multi-dimensional panoramic data; Performing space-time feature joint coding, comprising: Wherein, the For the time-coded vector to be a vector, For the spatial encoding of the vectors, For the expected trip time of the patient from the address to the hospital area, For the straight line distance of the patient from the address to the hospital area, Feature vectors are encoded for space-time union.
8. A dynamic optimization device for medical treatment resource reservation, comprising: The acquisition module is used for acquiring and preprocessing multidimensional panoramic data, wherein preprocessing comprises data cleaning, standardization and time-space feature joint coding; the patient clinical feature construction module is used for constructing a bimodal patient model of static clinical portraits and dynamic state evolution based on the preprocessed multidimensional panoramic data, and combining a dynamic weight clinical index evaluation and clinical priority dynamic quantification method to obtain clinical features of the patient; The constraint module is used for constructing a multi-dimensional coupling feature set based on clinical features of a patient, inputting the multi-dimensional coupling feature set into an examination duration prediction model with an online self-correction mechanism and based on a transducer attention mechanism to obtain an examination duration prediction result, and constructing an examination operation constraint matrix to obtain an examination operation constraint rule; The updating module is used for constructing a MADDPG multi-agent resource scheduling model under CTDE framework based on the checking duration prediction result and the checking operation constraint rule, and obtaining a resource scheduling joint optimization strategy through multi-agent joint training; And the scheme generating module is used for generating a reservation scheme based on the resource scheduling joint optimization strategy and carrying out rolling update on the reservation scheme based on a prospective rolling rescheduling mechanism.
9. An electronic device, comprising: A processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute to implement a method of dynamic optimization of medical care resource reservation as claimed in any one of claims 1 to 7.
10. A non-transitory computer readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform a method of implementing a dynamic optimization of medical care resource reservation as claimed in any one of claims 1 to 7.

Description

Dynamic optimization method, device, equipment and medium for medical treatment visit resource reservation Technical Field The invention relates to the field of diagnosis and treatment strategy optimization, in particular to a dynamic optimization method, device, equipment and medium for medical treatment and treatment resource reservation. Background With the continuous increase of medical service demands, medical treatment resources such as hospital image examination, examination and functional examination are increasingly tense, and patients often face the problems of long queuing time, unreasonable examination arrangement, low resource utilization rate and the like in the treatment process. Traditional medical appointment methods rely on manual experience or appointment systems based on simple rules, are generally arranged in a first-come first-appointment or fixed time slot allocation mode, lack of comprehensive consideration on clinical emergency degree of patients, difference of examination duration and real-time state of medical resources, and easily cause overlong waiting time of part of patients, and insufficient utilization of part of medical equipment or examination resources, so that overall medical service efficiency is affected. Therefore, how to improve the medical service efficiency and provide a more reasonable reservation mode is a problem to be solved. Disclosure of Invention The invention solves the technical problem of unreasonable reservation mode in the prior art by providing the dynamic optimization method, the device, the equipment and the medium for medical treatment visit resource reservation, and achieves the technical effect of improving the rationality of the reservation mode. In a first aspect, the present invention provides a method for dynamic optimization of medical care resource reservation, including: Acquiring and preprocessing multidimensional panoramic data, wherein preprocessing comprises data cleaning, standardization and time-space feature joint coding; Based on the preprocessed multidimensional panoramic data, constructing a bimodal patient model of static clinical portraits and dynamic state evolution, and combining a dynamic weight clinical index evaluation and clinical priority dynamic quantification method to obtain clinical characteristics of patients; Constructing a multi-dimensional coupling feature set based on clinical features of a patient, inputting the multi-dimensional coupling feature set into an examination duration prediction model with an online self-correction mechanism and based on a transducer attention mechanism to obtain an examination duration prediction result, and constructing an examination operation constraint matrix to obtain an examination operation constraint rule; Based on the checking duration prediction result and the checking operation constraint rule, constructing a MADDPG multi-agent resource scheduling model under CTDE framework, and obtaining a resource scheduling joint optimization strategy through multi-agent joint training; And generating a reservation scheme based on the resource scheduling combined optimization strategy, and carrying out rolling update on the reservation scheme based on a look-ahead rolling rescheduling mechanism. Further, based on the preprocessed multidimensional panoramic data, a bimodal patient model of static clinical portraits and dynamic state evolution is constructed, and a dynamic weighting clinical index evaluation and clinical priority dynamic quantification method is combined to obtain clinical characteristics of the patient, comprising: Constructing a static clinical portrayal model comprising: Wherein, the As a result of the static feature vector,For the feature vector of the basic information of the patient,Is the characteristic vector of the past diseases, allergies, complications and medical history of patients,Feature vectors are constrained for patient clinical paths and examination items,Reserving a performance rate and checking fit characteristic vector for patient history; constructing a dynamic state evolution model, comprising: based on real-time physiological index data, constructing time sequence dynamic feature vector ; Based on time sequence dynamic characteristic vectorOutputting a dynamic state evolution model, comprising: Wherein, the Is thatThe hidden layer state of the LSTM network at the moment,In order to be the hidden layer state at the previous moment,For the weight matrix of the LSTM network,Is thatThe dynamic state vector of the patient at the moment,The function is activated for Sigmoid,Is a weight matrix of the full connection layer,Bias terms for the full connection layer; A dynamic weighted clinical index assessment system, wherein the dynamic weighted clinical index assessment system is associated with a treatment stage and a disease type; under the dynamic weight clinical index evaluation system, carrying out clinical priority dynamic quantification, which co