Business model based on public hospital operation and management scenario

Abstract

The present invention discloses a business model based on a public hospital operation and management scenario, and relates to the field of business management technology. The business model based on the public hospital operation and management scenario includes: a business data collection module, a business data analysis module and a business data feedback module. The present invention uses periodic data collection and feature analysis, uses XGBoost and Informer models to screen key features, and combines the LSTM model improved by dynamic multi-fractal wavelet decomposition for predictive analysis, so as to effectively identify the periodic fluctuation characteristics and control anomalies of the consumables business of public hospitals; uses the prediction results of the model to guide the optimization of operation and management, improves the efficiency of consumables use and cost control; at the same time, through the business data feedback module, the model can self-adjust and optimize according to the actual control data, enhance the prediction accuracy and adaptability, and realize the refined management of the consumables business of public hospitals.

Description

Technical Field

[0001] The present invention relates to the field of business management technology, and in particular to a business model based on a public hospital operation and management scenario. Background Art

[0002] In the operation and management scenario of public hospitals, the purchase of consumables is strictly controlled. Consumables are an important expenditure item in hospital operations. Strict control can reduce unnecessary purchases and waste, help hospitals optimize budgets and reduce costs. In the existing technology, consumables are generally managed based on the data of the consumables themselves, and consumables are identified using RFID tags. The storage location, inventory status, consumption, etc. of the consumables can be tracked in real time. However, RFID tags are mainly used to track and manage the physical status and flow data of a single consumable, but they themselves cannot handle the integration with other related data. It is difficult for hospitals to conduct a comprehensive global analysis based on a single consumable data, and it is impossible to conduct a comprehensive assessment of the entire supply chain, financial status or the overall consumable usage of the hospital.

[0003] Therefore, it is necessary to conduct a global analysis based on the business data of public hospitals. The analysis of historical data and usage trends can accurately predict future consumables demand, avoid over-purchasing or insufficient inventory, and ensure efficient allocation of resources. Summary of the invention

[0004] The present invention aims to provide a business model based on the public hospital operation and management scenario, and conduct a global feature analysis of the consumables business to optimize the consumables management and control in the next stage.

[0005] The business model based on the public hospital operation and management scenario includes:

[0006] A business data collection module, including a cycle setting unit and a data collection unit;

[0007] The cycle setting unit is used to set the first analysis cycle, the second analysis cycle and the third analysis cycle;

[0008] The data collection unit is used to obtain the consumables business data set X, X according to the first analysis cycle i ={X i |i=1,2,…,I};X i represents the i-th business data in the consumables business data set X, and I is the total number of business data in the consumables business data set X; the consumables business data set X is input into the business data feature analysis model for analysis to obtain the consumables business data periodic fluctuation feature T x ;

[0009] The business data feature analysis model includes a data preprocessing layer, a feature correlation analysis layer, and a feature fusion output layer. It is constructed by fusing the XGBoost and Informer models to perform feature screening on the consumables business data set X.

[0010] Business data analysis module, including feature analysis unit and data optimization unit;

[0011] The characteristic analysis unit is used to obtain the periodic fluctuation characteristics of all consumable business data in the second analysis period, and obtain the periodic fluctuation characteristics T of the consumable business data. xn , n=1, 2, ..., N, N is the total number of periodic fluctuation features of all consumable business data in the second analysis period; all consumable business data periodic fluctuation features T xn The data is input into the business data evaluation and prediction model for analysis to obtain the consumables business control prediction strategy and consumables business control abnormal information; the business data evaluation and prediction model includes a feature connection layer, a feature judgment layer, a prediction analysis layer and a result output layer. The LSTM model is improved by using dynamic multi-fractal wavelet decomposition to analyze the periodic fluctuation characteristics T of all consumables business data. xn Conduct predictive analysis and judgment;

[0012] The data optimization unit is used to optimize the operation and management of public hospitals based on the abnormal information of consumable business control and the prediction strategy of consumable business control;

[0013] A business data feedback module, including a feedback adjustment unit;

[0014] The feedback adjustment unit is used to obtain the actual management and control data of all consumables businesses within the third analysis cycle; based on the actual management and control data of all consumables businesses and the business data feedback adjustment model, the consumables business data feedback results are obtained, and the business data evaluation and prediction model is updated according to the consumables business data feedback results; the business data feedback adjustment model includes a feedback feature evaluation layer, a feedback result quantitative analysis layer and a model adjustment layer.

[0015] As a preferred technical solution of the present invention, the data preprocessing layer is used to process the business data X i Perform preprocessing to obtain preprocessed business data X i ';

[0016] The feature correlation analysis layer is used to analyze all pre-processed business data X i 'Perform feature correlation analysis to obtain feature correlation screening results; based on the feature correlation screening results, perform feature correlation screening on all pre-processed business data X i 'Screen and obtain pre-screened business data Y m , m=1, 2, ..., M; M is the total number of pre-processed service data X i'The total number of data retained after filtering;

[0017] The feature fusion output layer is used to convert all pre-screened business data Y m Perform feature fusion to obtain the periodic fluctuation feature T of consumable business data x .

[0018] As a preferred technical solution of the present invention, the specific steps of performing feature correlation analysis on the pre-processed business data Xi' in the feature correlation analysis layer include:

[0019] The feature correlation analysis layer includes a feature screening layer, a rough classification layer, and a correction layer; the rough classification layer is trained based on the XGBoost model, and the correction layer is constructed based on the Informer model;

[0020] In the feature screening layer, the preprocessed business data X i 'Perform feature extraction to obtain preprocessed business data feature Z i ; Calculate the preprocessing business data feature Z i SHAP value of the business data contribution G i ; The business data contribution G i Compare with the business data contribution threshold, and calculate the contribution of all business data G i Preprocessed business data features Z greater than the business data contribution threshold i Input into the coarse classification layer; otherwise, stop data input;

[0021] In the coarse classification layer, the contribution G to all business data i Preprocessed business data features Z greater than the business data contribution threshold i Perform rough classification to obtain rough classification feature results of business data;

[0022] In the correction layer, the rough classification feature results of business data are corrected to obtain the feature correlation screening results.

[0023] As a preferred technical solution of the present invention, the specific steps of training the coarse classification layer and the correction layer include:

[0024] Collecting several groups of coarse classification training samples; each group of coarse classification training samples contains feature data and corresponding business feature correlation; combining several groups of coarse classification training samples to obtain a coarse classification layer training set;

[0025] Input the coarse classification layer training set into the XGBoost model for training to obtain the initial coarse classification layer; perform model evaluation on the initial coarse classification layer. If the initial coarse classification layer passes the model evaluation, the initial classification layer is used as the coarse classification layer in the feature association analysis layer; otherwise, continue model training using the coarse classification layer training set;

[0026] During the training of the coarse classification layer, a new decision tree is added in each training, and the error value of the previous training is used as the key correction value of the next training; after multiple rounds of iterative training, the initial coarse classification layer is obtained;

[0027] Collecting several groups of timing correction training samples; each group of timing correction training samples contains feature data and corresponding service feature timing labels; combining several groups of timing correction training samples to obtain a correction layer training set;

[0028] The correction layer training set is input into the Informer model for training to obtain the initial correction layer; the divergence index is used as the model evaluation index of the initial correction layer; the divergence index of the initial correction layer is calculated to obtain the divergence index of the initial correction layer; if the divergence index of the initial correction layer meets the preset model evaluation value, the initial correction layer is used as the correction layer in the feature correlation analysis layer; otherwise, the model training is continued using the correction layer training set.

[0029] As a preferred technical solution of the present invention,

[0030] The feature connection layer is used to transform all the periodic fluctuation characteristics of consumable business data into xn Combine them to get the periodic fluctuation feature set T of consumable business data xn ';

[0031] The feature judgment layer is used to identify the periodic fluctuation feature set T of the consumables business data. xn 'Make abnormal judgment and obtain abnormal information of consumable business management and control;

[0032] The prediction analysis layer is used to analyze the periodic fluctuation feature set T of consumable business data. xn 'Perform forecast analysis to obtain the forecast strategy for the consumables business management and control in the next second analysis cycle;

[0033] The result output layer is used to output the abnormal information of consumable business control and the prediction strategy of consumable business control.

[0034] As a preferred technical solution of the present invention, the consumables business data period fluctuation feature set T is analyzed in the prediction analysis layer. xn 'The specific steps for predictive analysis include:

[0035] The prediction analysis layer is built based on the LSTM model, which contains a trained prediction network. xn 'Perform dynamic multi-fractal wavelet decomposition to obtain the optimized value H of the periodic fluctuation characteristics of consumable business data xn And the optimization sequence L of the periodic fluctuation characteristics of consumable business data xn ;

[0036] Optimizing the value H based on the periodic fluctuation characteristics of consumables business data xn Optimize sequence L for periodic fluctuation characteristics of consumable business data xn Cutting is performed to obtain the optimized sequence L of the periodic fluctuation characteristics of the cutting consumables business data xn ';

[0037] Optimize the sequence L for the periodic fluctuation characteristics of cutting consumables business data xn 'Input into the prediction network in the prediction analysis layer for analysis to obtain the predicted consumables business data sequence C xn ; Based on the forecast consumables business data sequence C xn Determine the forecast strategy for the consumables business management and control within the next second analysis cycle.

[0038] As a preferred technical solution of the present invention, the feedback feature evaluation layer is used to perform feature evaluation on all actual control data of consumables business to obtain actual control evaluation value of consumables business; based on the actual control evaluation value of consumables business, judgment is made to obtain the feedback result of consumables business data;

[0039] The feedback result quantification analysis layer is used to quantify feedback features based on the actual management and control evaluation value of the consumables business and the consumables business data feedback results to obtain the hyperparameters for adjusting the prediction model;

[0040] The model adjustment layer is used to update the business data evaluation prediction model based on the prediction model by adjusting the hyperparameters.

[0041] As a preferred technical solution of the present invention, the training feedback result quantization analysis layer specifically comprises the following steps:

[0042] Based on the actual management and control evaluation value of consumables business, K models are constructed to adjust the parameter individual J k ; Each model adjusts the parameters of individual J k It contains the hyperparameter values ​​for adjusting the number of neurons in the three hidden layers of the prediction network, the loss rate, and the number of training samples; the model adjustment parameter individual J k Combine to obtain the iterative population of model adjustment parameters; set the maximum number of iterations;

[0043] In the population iteration process, the gray wolf optimization algorithm is used to iterate the model adjustment parameter iterative population;

[0044] Based on the model adjustment parameters individual J k The business data evaluation prediction model is simulated and adjusted to obtain a simulated business data evaluation prediction model; the mean square error of the simulated business data evaluation prediction model during simulation prediction is calculated, and the mean square error is used as the model adjustment parameter individual J k The fitness S k ;

[0045] When the maximum number of iterations is reached, the model adjustment parameter individual corresponding to the maximum output fitness is the optimal model adjustment parameter individual; based on the optimal model adjustment parameter individual, the prediction model adjustment hyperparameters are obtained.

[0046] The present invention has the following advantages:

[0047] The present invention uses XGBoost and Informer models to screen key features through periodic data collection and feature analysis, and combines the LSTM model improved by dynamic multifractal wavelet decomposition for predictive analysis, so as to effectively identify the periodic fluctuation characteristics and management and control anomalies of the consumables business of public hospitals; the prediction results of the model are used to guide the optimization of operation management, improve the efficiency of consumables use and cost control; at the same time, through the business data feedback module, the model can self-adjust and optimize according to the actual management and control data, enhance the prediction accuracy and adaptability, and realize the refined management of the consumables business of public hospitals.

[0048] The present invention combines XGBoost and Informer models in the feature correlation analysis layer to achieve efficient feature screening and correlation analysis of pre-processed business data; calculates the contribution of business data through SHAP value to screen out features that have a significant impact on the business; roughly classifies the screened features and performs time series correction on the classification results, thereby improving the accuracy of feature classification and the accuracy of time series prediction. This method of combining rough classification and time series correction not only enhances the model's ability to identify the importance of features, but also optimizes the analysis results of feature correlation, providing strong data support for the refined management and prediction of consumables business in public hospitals. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a schematic diagram of the structure of a business model based on a public hospital operation and management scenario adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0051] For business models based on public hospital operation and management scenarios, see Figure 1As shown, including:

[0052] A business data collection module, including a cycle setting unit and a data collection unit;

[0053] The cycle setting unit is used to set the first analysis cycle, the second analysis cycle and the third analysis cycle; the length of the third analysis cycle is greater than the second analysis cycle, and the length of the second analysis cycle is greater than the first analysis cycle; the length units of the first analysis cycle, the first analysis cycle and the first analysis cycle are all days; the first analysis cycle, the second analysis cycle and the third analysis cycle are set by professional and technical personnel according to actual conditions;

[0054] The data collection unit is used to obtain the consumables business data set X, X according to the first analysis cycle i ={X i |i=1,2,…,I};X i represents the i-th business data in the consumables business data set X, and I is the total number of business data in the consumables business data set X; the consumables business data set X is input into the business data feature analysis model for analysis to obtain the consumables business data periodic fluctuation feature T x ;

[0055] The business data feature analysis model includes a data preprocessing layer, a feature correlation analysis layer, and a feature fusion output layer. It is constructed by fusing the XGBoost and Informer models to perform feature screening on the consumables business data set X.

[0056] The data preprocessing layer is used to process business data X i Perform preprocessing to obtain preprocessed business data X i '; By preprocessing the original business data, such as denoising, missing value filling, standardization and other operations, the quality of the data can be improved, unnecessary noise and inconsistency can be eliminated, and the data can be more suitable for subsequent analysis; the business data obtained after preprocessing will help improve the accuracy and stability of feature analysis;

[0057] The feature correlation analysis layer is used to analyze all pre-processed business data X i 'Perform feature correlation analysis to obtain feature correlation screening results; based on the feature correlation screening results, perform feature correlation screening on all pre-processed business data X i 'Screen and obtain pre-screened business data Y m , m=1, 2, ..., M; M is the total number of pre-processed service data X i'The total number of data retained after screening; by analyzing the correlation between each feature, it can automatically identify which features are most critical to the target consumables task; feature correlation analysis helps the model filter out the most relevant features, eliminate redundant or irrelevant features, and avoid the negative impact of too many invalid features on model performance;

[0058] For example, business data may include changes in the quantity and type of consumables, types of consumable expenditures in hospital departments, and patient return visit rates in hospital departments. These business data should be screened for relevance to retain the data most relevant to consumables management and control.

[0059] The feature fusion output layer is used to convert all pre-screened business data Y m Perform feature fusion to obtain the periodic fluctuation feature T of consumable business data x ; By fusing the pre-screened business data, the impact of multiple features on the periodic fluctuation characteristics can be integrated, thereby improving the accuracy of the prediction model; the role of the feature fusion layer is to ensure the synergy of multiple information, reduce the error that may be caused by a single feature, and improve the overall model's ability to capture periodic fluctuation characteristics;

[0060] The specific steps of performing feature correlation analysis on the pre-processed business data Xi' in the feature correlation analysis layer include:

[0061] The feature correlation analysis layer includes a feature screening layer, a rough classification layer, and a correction layer; the rough classification layer is trained based on the XGBoost model, and the correction layer is constructed based on the Informer model;

[0062] In the feature screening layer, the preprocessed business data X i 'Perform feature extraction to obtain preprocessed business data feature Z i ; Calculate the preprocessing business data feature Z i SHAP value of the business data contribution G i ; The business data contribution G i Compare with the business data contribution threshold, and calculate the contribution of all business data G i Preprocessed business data features Z greater than the business data contribution threshold i Input into the coarse classification layer; otherwise, stop data input;

[0063] The feature screening layer calculates the importance of business data features through SHAP values, which can accurately quantify the contribution of each feature to model prediction. This quantified contribution helps to automatically screen out the features that have the greatest impact on the target task, while irrelevant or less contributing features will be eliminated, thereby improving data quality and model efficiency.

[0064] In the coarse classification layer, the contribution G to all business data i Preprocessed business data features Z greater than the business data contribution threshold i Perform rough classification to obtain rough classification feature results of business data; train based on the XGBoost model, so that the screened business data features can be roughly classified and the data can be grouped or marked according to different categories; XGBoost can effectively handle nonlinear relationships and high-dimensional data, and improve the accuracy of classification through multiple iterations; through the rough classification process, the data can be divided into different groups, providing structured information for subsequent correction and feature analysis, making it easier for the model to extract useful patterns from different categories;

[0065] In the correction layer, the rough classification feature results of business data are corrected to obtain the feature correlation screening results; it is built based on the Informer model, which is a model specifically used to process time series data and can capture and correct long-term dependencies in time series; Informer is particularly suitable for processing data containing time series features, which enables the model to take into account the time series patterns in business data, thereby further improving the accuracy of feature analysis.

[0066] The specific steps for training the coarse classification layer and the correction layer include:

[0067] Collecting several groups of coarse classification training samples; each group of coarse classification training samples contains feature data and corresponding business feature correlation; combining several groups of coarse classification training samples to obtain a coarse classification layer training set;

[0068] Input the coarse classification layer training set into the XGBoost model for training to obtain the initial coarse classification layer; perform model evaluation on the initial coarse classification layer. If the initial coarse classification layer passes the model evaluation, the initial classification layer is used as the coarse classification layer in the feature association analysis layer; otherwise, continue model training using the coarse classification layer training set;

[0069] During the training of the coarse classification layer, a new decision tree is added in each training, and the error value of the previous training is used as the key correction value of the next training; after multiple rounds of iterative training, the initial coarse classification layer is obtained;

[0070] Collecting several groups of timing correction training samples; each group of timing correction training samples contains feature data and corresponding service feature timing labels; combining several groups of timing correction training samples to obtain a correction layer training set;

[0071] Input the correction layer training set into the Informer model for training to obtain the initial correction layer; use the divergence index as the model evaluation index of the initial correction layer; calculate the divergence index of the initial correction layer to obtain the divergence index of the initial correction layer; if the divergence index of the initial correction layer meets the preset model evaluation value, the initial correction layer is used as the correction layer in the feature correlation analysis layer; otherwise, continue model training using the correction layer training set;

[0072] Through multiple rounds of training and evaluation, the coarse classification layer and the correction layer can be adjusted and optimized based on feedback. Each round of training adds a new decision tree to correct errors, ensuring that the model gradually converges and reducing the risk of overfitting during training. The optimized coarse classification layer and correction layer can effectively improve the accuracy and generalization ability of the model; through the feature correlation analysis of consumable business data and the accurate identification of periodic fluctuations, hospital managers can make more scientific decisions based on accurate prediction results;

[0073] Business data analysis module, including feature analysis unit and data optimization unit;

[0074] The characteristic analysis unit is used to obtain the periodic fluctuation characteristics of all consumable business data in the second analysis period, and obtain the periodic fluctuation characteristics T of the consumable business data. xn , n=1, 2, ..., N, N is the total number of periodic fluctuation features of all consumable business data in the second analysis period; all consumable business data periodic fluctuation features T xn The data is input into the business data evaluation and prediction model for analysis to obtain the consumables business control prediction strategy and consumables business control abnormal information; the business data evaluation and prediction model includes a feature connection layer, a feature judgment layer, a prediction analysis layer and a result output layer. The LSTM model is improved by using dynamic multi-fractal wavelet decomposition to analyze the periodic fluctuation characteristics T of all consumables business data. xn Conduct predictive analysis and judgment;

[0075] The feature connection layer is used to transform all the periodic fluctuation characteristics of consumable business data into xn Combine them to get the periodic fluctuation feature set T of consumable business data xn ';

[0076] The feature judgment layer is used to identify the periodic fluctuation feature set T of the consumables business data. xn 'Make abnormal judgment and obtain abnormal information of consumable business management and control;

[0077] The prediction analysis layer is used to analyze the periodic fluctuation feature set T of consumable business data. xn 'Perform forecast analysis to obtain the forecast strategy for the consumables business management and control in the next second analysis cycle;

[0078] The result output layer is used to output the abnormal information of consumable business control and the prediction strategy of consumable business control;

[0079] The prediction analysis layer uses the LSTM model improved by dynamic multi-fractal wavelet decomposition to capture the complex periodicity and trend characteristics in the consumables data and accurately predict the future demand for consumables. This prediction analysis can help hospital managers understand the future changes in demand for consumables in advance, formulate scientific procurement and inventory plans, and avoid resource waste and insufficient inventory. The feature judgment layer makes real-time abnormal judgments on consumables data, which helps to timely identify abnormal patterns or fluctuations in the data. These anomalies may come from unexpected consumables consumption, improper procurement, or management errors.

[0080] In the prediction analysis layer, the periodic fluctuation feature set T of the consumables business data is xn 'The specific steps for predictive analysis include:

[0081] The prediction analysis layer is built based on the LSTM model, which contains a trained prediction network. xn 'Perform dynamic multi-fractal wavelet decomposition to obtain the optimized value H of the periodic fluctuation characteristics of consumable business data xn And the optimization sequence L of the periodic fluctuation characteristics of consumable business data xn ;

[0082] Optimizing the value H based on the periodic fluctuation characteristics of consumables business data xn Optimize sequence L for periodic fluctuation characteristics of consumable business data xn Cutting is performed to obtain the optimized sequence L of the periodic fluctuation characteristics of the cutting consumables business data xn ';

[0083] Optimize the sequence L for the periodic fluctuation characteristics of cutting consumables business data xn 'Input into the prediction network in the prediction analysis layer for analysis to obtain the predicted consumables business data sequence C xn ; Based on the forecast consumables business data sequence C xn Determine the forecast strategy for the consumables business management and control in the next second analysis cycle;

[0084] The introduction of dynamic multi-fractal wavelet decomposition can effectively decompose the fluctuation characteristics of complex time series and extract the multi-level periodic fluctuation characteristics of consumables data by improving the LSTM model. This improved method can more accurately characterize the complex periodic changes of data, so that the model can better adapt to the analysis of multi-period and multi-frequency consumables data. In actual operation, consumables data is often affected by sudden demand or policy changes. Through the abnormal identification of the feature judgment layer and the strategy adjustment of the prediction analysis layer, the model can flexibly adapt to new situations and maintain high prediction performance under unstable market conditions.

[0085] The data optimization unit is used to optimize the operation and management of public hospitals based on the abnormal information of consumable business control and the prediction strategy of consumable business control;

[0086] A business data feedback module, including a feedback adjustment unit;

[0087] The feedback adjustment unit is used to obtain the actual control data of all consumables businesses in the third analysis cycle; based on the actual control data of all consumables businesses and the business data feedback adjustment model, the consumables business data feedback results are obtained, and the business data evaluation and prediction model is updated according to the consumables business data feedback results; the business data feedback adjustment model includes a feedback feature evaluation layer, a feedback result quantitative analysis layer and a model adjustment layer;

[0088] The feedback feature evaluation layer is used to perform feature evaluation on all actual control data of consumables business to obtain the actual control evaluation value of consumables business; make judgments based on the actual control evaluation value of consumables business to obtain the feedback result of consumables business data;

[0089] The feedback result quantification analysis layer is used to quantify feedback features based on the actual management and control evaluation value of the consumables business and the consumables business data feedback results to obtain the hyperparameters for adjusting the prediction model;

[0090] The model adjustment layer is used to adjust the hyperparameters of the prediction model to update the business data evaluation prediction model;

[0091] Training feedback result quantitative analysis layer, the specific steps include:

[0092] Based on the actual management and control evaluation value of consumables business, K models are constructed to adjust the parameter individual J k ; Each model adjusts the parameters of individual J k It contains the hyperparameter values ​​for adjusting the number of neurons in the three hidden layers of the prediction network, the loss rate, and the number of training samples; the model adjustment parameter individual J k Combination, to obtain the iterative population of model adjustment parameters; set the maximum number of iterations; the maximum number of iterations is set by professional technicians according to actual conditions;

[0093] In the population iteration process, the gray wolf optimization algorithm is used to iterate the model adjustment parameter iterative population;

[0094] Based on the model adjustment parameters individual J k The business data evaluation prediction model is simulated and adjusted to obtain a simulated business data evaluation prediction model; the mean square error of the simulated business data evaluation prediction model during simulation prediction is calculated, and the mean square error is used as the model adjustment parameter individual J k The fitness S k ;

[0095] When the maximum number of iterations is reached, the model adjustment parameter individual corresponding to the maximum fitness is output, which is the optimal model adjustment parameter individual; based on the optimal model adjustment parameter individual, the prediction model adjustment hyperparameter is obtained;

[0096] The feedback feature evaluation layer obtains actual control data, evaluates and feedbacks the actual situation of consumables management, and identifies the gap between model prediction and actual situation by comparing with predicted data. This closed-loop feedback mechanism helps the model to continuously adjust according to the actual effect, ensuring that the prediction results are more consistent with actual needs; the feedback result quantitative analysis layer generates quantitative feedback features based on the evaluation results of actual data, determines the adjustment direction and adjustment range of the prediction model, and generates the adjustment hyperparameters of the model; through this data-driven adaptive adjustment mechanism, the model can automatically optimize key parameters, such as the number of hidden layer neurons and loss rate, so that it can adapt to changing business needs and data patterns; the gray wolf optimization algorithm intelligently optimizes the individual model adjustment parameters during the iteration process to help determine the optimal model hyperparameter settings; the gray wolf optimization algorithm simulates the hunting behavior of wolves to perform global and local searches, which can avoid falling into the local optimal solution, thereby obtaining the parameter combination with the highest fitness; based on the self-adjustment and accuracy improvement of the prediction model, the hospital can manage consumables needs more efficiently and achieve more accurate procurement and inventory management. This refined management strategy helps hospitals optimize resource allocation, reduce resource waste, and reduce operating costs while meeting clinical needs.

[0097] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. The business system based on the public hospital operation and management scenario is characterized by: include: A business data collection module, including a cycle setting unit and a data collection unit; The cycle setting unit is used to set the first analysis cycle, the second analysis cycle and the third analysis cycle; The data collection unit is used to obtain the consumables business data set X, X according to the first analysis cycle i ={X i |i=1,2,…,I};X i represents the i-th business data in the consumables business data set X, and I is the total number of business data in the consumables business data set X; the consumables business data set X is input into the business data feature analysis model for analysis to obtain the consumables business data periodic fluctuation feature T x ; The business data feature analysis model includes a data preprocessing layer, a feature correlation analysis layer, and a feature fusion output layer. It is constructed by fusing the XGBoost and Informer models to perform feature screening on the consumables business data set X. Business data analysis module, including feature analysis unit and data optimization unit; The characteristic analysis unit is used to obtain the periodic fluctuation characteristics of all consumable business data in the second analysis period, and obtain the periodic fluctuation characteristics T of the consumable business data. xn , n=1, 2, ..., N, where N is the total number of periodic fluctuation characteristics of all consumables business data in the second analysis period; All consumables business data periodic fluctuation characteristics T xn The data is input into the business data evaluation and prediction model for analysis to obtain the consumables business control prediction strategy and consumables business control abnormal information; the business data evaluation and prediction model includes a feature connection layer, a feature judgment layer, a prediction analysis layer and a result output layer. The LSTM model is improved by using dynamic multi-fractal wavelet decomposition to analyze the periodic fluctuation characteristics T of all consumables business data. xn Conduct predictive analysis and judgment; The data optimization unit is used to optimize the operation and management of public hospitals based on the abnormal information of consumable business control and the prediction strategy of consumable business control; A business data feedback module, including a feedback adjustment unit; The feedback adjustment unit is used to obtain the actual management and control data of all consumables businesses within the third analysis cycle; based on the actual management and control data of all consumables businesses and the business data feedback adjustment model, the consumables business data feedback results are obtained, and the business data evaluation and prediction model is updated according to the consumables business data feedback results; the business data feedback adjustment model includes a feedback feature evaluation layer, a feedback result quantitative analysis layer and a model adjustment layer.

2. The business system based on the public hospital operation and management scenario according to claim 1 is characterized in that: The data preprocessing layer is used to process business data X i Perform preprocessing to obtain preprocessed business data X i '; The feature correlation analysis layer is used to analyze all pre-processed business data X i 'Perform feature correlation analysis to obtain feature correlation screening results; Based on the feature correlation screening results, all pre-processed business data X i 'Screen and get pre-screened business data Y m , m=1,2,…,M; M is the sum of all pre-processed business data X i 'The total number of data retained after filtering; The feature fusion output layer is used to convert all pre-screened business data Y m Perform feature fusion to obtain the periodic fluctuation feature T of consumable business data x .

3. The business system based on the public hospital operation and management scenario according to claim 2 is characterized in that: The specific steps of performing feature correlation analysis on the pre-processed business data Xi' in the feature correlation analysis layer include: The feature correlation analysis layer includes a feature screening layer, a rough classification layer, and a correction layer; the rough classification layer is trained based on the XGBoost model, and the correction layer is constructed based on the Informer model; In the feature screening layer, the preprocessed business data X i 'Perform feature extraction to obtain preprocessed business data feature Z i ; Calculate the preprocessing business data feature Z i SHAP value of the business data contribution G i ; The business data contribution G i Compare with the business data contribution threshold, and calculate the contribution of all business data G i Preprocessed business data features Z greater than the business data contribution threshold i Input into the coarse classification layer; otherwise, stop data input; In the coarse classification layer, the contribution G to all business data i Preprocessed business data features Z greater than the business data contribution threshold i Perform rough classification to obtain rough classification feature results of business data; In the correction layer, the rough classification feature results of business data are corrected to obtain the feature correlation screening results.

4. The business system based on the public hospital operation and management scenario according to claim 3 is characterized in that: The specific steps for training the coarse classification layer and the correction layer include: Collecting several groups of coarse classification training samples; each group of coarse classification training samples contains feature data and corresponding business feature correlation; combining several groups of coarse classification training samples to obtain a coarse classification layer training set; Input the coarse classification layer training set into the XGBoost model for training to obtain the initial coarse classification layer; perform model evaluation on the initial coarse classification layer. If the initial coarse classification layer passes the model evaluation, the initial classification layer is used as the coarse classification layer in the feature association analysis layer; otherwise, continue model training using the coarse classification layer training set; During the training of the coarse classification layer, a new decision tree is added in each training, and the error value of the previous training is used as the key correction value of the next training; after multiple rounds of iterative training, the initial coarse classification layer is obtained; Collecting several groups of timing correction training samples; each group of timing correction training samples contains feature data and corresponding service feature timing labels; combining several groups of timing correction training samples to obtain a correction layer training set; The correction layer training set is input into the Informer model for training to obtain the initial correction layer; the divergence index is used as the model evaluation index of the initial correction layer; the divergence index of the initial correction layer is calculated to obtain the divergence index of the initial correction layer; if the divergence index of the initial correction layer meets the preset model evaluation value, the initial correction layer is used as the correction layer in the feature correlation analysis layer; otherwise, the model training is continued using the correction layer training set.

5. The business system based on the public hospital operation and management scenario according to claim 4 is characterized in that: The feature connection layer is used to transform all the periodic fluctuation characteristics of consumable business data into xn Combine them to get the periodic fluctuation feature set T of consumable business data xn '; The feature judgment layer is used to identify the periodic fluctuation feature set T of the consumables business data. xn 'Make abnormal judgment and obtain abnormal information of consumable business management and control; The prediction analysis layer is used to analyze the periodic fluctuation feature set T of consumable business data. xn 'Perform forecast analysis to obtain the forecast strategy for the consumables business management and control in the next second analysis cycle; The result output layer is used to output the abnormal information of consumable business control and the prediction strategy of consumable business control.

6. The business system based on the public hospital operation and management scenario according to claim 5 is characterized in that: In the prediction analysis layer, the periodic fluctuation feature set T of the consumables business data is xn 'The specific steps for predictive analysis include: The prediction analysis layer is built based on the LSTM model, which contains a trained prediction network. xn 'Perform dynamic multi-fractal wavelet decomposition to obtain the optimized value H of the periodic fluctuation characteristics of consumable business data xn And the optimization sequence L of the periodic fluctuation characteristics of consumable business data xn ; Optimizing the value H based on the periodic fluctuation characteristics of consumables business data xn Optimize sequence L for periodic fluctuation characteristics of consumable business data xn Cutting is performed to obtain the optimized sequence L of the periodic fluctuation characteristics of the cutting consumables business data xn '; Optimize the sequence L for the periodic fluctuation characteristics of cutting consumables business data xn 'Input into the prediction network in the prediction analysis layer for analysis to obtain the predicted consumables business data sequence C xn ; Based on the forecast consumables business data sequence C xn Determine the forecast strategy for the consumables business management and control within the next second analysis cycle.

7. The business system based on the public hospital operation and management scenario according to claim 6 is characterized in that: The feedback feature evaluation layer is used to perform feature evaluation on all actual control data of consumables business to obtain the actual control evaluation value of consumables business; make judgments based on the actual control evaluation value of consumables business to obtain the feedback result of consumables business data; The feedback result quantification analysis layer is used to quantify feedback features based on the actual management and control evaluation value of the consumables business and the consumables business data feedback results to obtain the hyperparameters for adjusting the prediction model; The model adjustment layer is used to update the business data evaluation prediction model based on the prediction model by adjusting the hyperparameters.

8. The business system based on the public hospital operation and management scenario according to claim 7 is characterized in that: Training feedback result quantitative analysis layer, the specific steps include: Based on the actual management and control evaluation value of consumables business, K models are constructed to adjust the parameter individual J k ; Each model adjusts the parameters of individual J k It contains the hyperparameter values ​​for adjusting the number of neurons in the three hidden layers of the prediction network, the loss rate, and the number of training samples; the model adjustment parameter individual J k Combine to obtain the iterative population of model adjustment parameters; set the maximum number of iterations; In the population iteration process, the gray wolf optimization algorithm is used to iterate the model adjustment parameter iterative population; Based on the model adjustment parameters individual J k The business data evaluation prediction model is simulated and adjusted to obtain a simulated business data evaluation prediction model; the mean square error of the simulated business data evaluation prediction model during simulation prediction is calculated, and the mean square error is used as the model adjustment parameter individual J k The fitness S k ; When the maximum number of iterations is reached, the model adjustment parameter individual corresponding to the maximum output fitness is the optimal model adjustment parameter individual; based on the optimal model adjustment parameter individual, the prediction model adjustment hyperparameters are obtained.

Images