Risk early warning system

Abstract

The embodiment of the invention provides a risk early warning system, which comprises a cloud system and an edge system, and is characterized in that the cloud system is used for collecting and processing clinical data of different patients, and training a large neural network model based on the collected and processed data, so as to obtain a complex model capable of carrying out assessment and risk early warning on different clinical manifestations of the patients; then distilling the complex model by utilizing a knowledge distillation technology to obtain a small efficient model; and the edge system is used for deploying the small efficient models on each edge device so as to realize clinical risk monitoring and early warning by using the small efficient models, and the edge system is also used for performing data-related synchronous processing on the complex model and the small efficient models. According to the risk early warning system, the clinical reasoning efficiency and the clinical response speed can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of risk early warning, and in particular relates to a risk early warning system. Background Art

[0002] The risk warning model in the clinical field has certain requirements on the reasoning efficiency and response speed of the model. The traditional risk warning model cannot adapt to the reasoning efficiency and response speed requirements of the clinical field.

[0003] In the Chinese Public Invention with application number CN202210939845.1, a perioperative risk warning method based on machine learning is disclosed, which obtains evaluation parameters associated with perioperative risk assessment of patients to be operated on; collects multiple first data corresponding to the evaluation parameters of patients to be operated on; processes multiple first data, inputs the obtained processing results into the perioperative risk prediction model, and outputs the perioperative risk warning results corresponding to the patients to be operated on. This method can automatically predict and warn various risks that may occur during the perioperative period based on the evaluation parameters and corresponding data associated with the perioperative risk assessment, through the perioperative risk prediction model. This method does not need to rely on manual experience, and can detect perioperative adverse events as early as possible, thereby effectively reducing postoperative complications and ensuring surgical safety.

[0004] The defect of the existing technology is that although it can effectively reduce postoperative complications and ensure surgical safety, it lacks training and tuning of the prediction model, resulting in limited reasoning efficiency and response speed of the model, which cannot adapt to high clinical reasoning efficiency and fast response speed. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a risk warning system that can improve clinical reasoning efficiency and clinical response speed.

[0006] The content of the embodiment of the present invention includes providing a risk warning system, including a cloud system and an edge system, wherein the cloud system is used to collect and process clinical data of different patients, and train a large neural network model based on the collected and processed data to obtain a complex model that can evaluate different clinical manifestations of patients and provide risk warnings, and then use knowledge distillation technology to distill the complex model into a small and efficient model;

[0007] The edge system is used to deploy the small and efficient model on various edge devices to utilize the small and efficient model to achieve clinical risk monitoring and early warning. The edge system is also used to synchronize data of the complex model and the small and efficient model.

[0008] In some embodiments, the cloud system includes:

[0009] The data collection and preprocessing module is used to obtain vital sign data, electronic medical record data, and expert annotation data of different patients in different clinical scenarios, and clean and standardize the obtained data;

[0010] A model training module, used to train a large neural network model using preprocessed data to obtain a complex model, wherein the complex model is used to perform patient emergency classification, vital sign risk warning, intraoperative hypotension and hypoxemia risk warning;

[0011] A model optimization module, used for optimizing the complex model;

[0012] The knowledge distillation module is used to transfer the knowledge of the optimized complex model through the knowledge distillation technology to obtain a small and efficient model.

[0013] In some embodiments, the edge system comprises:

[0014] An edge model deployment and reasoning module, which is used to deploy the small and efficient model to various edge devices at the clinical site to monitor and warn patients of risks based on the clinical data in the edge devices;

[0015] A model update and synchronization mechanism module, used to update the models on the cloud system and edge system using an incremental update method;

[0016] The data flow and communication protocol module is used to implement data exchange between various models in the cloud system and the edge system using the HTTP communication protocol;

[0017] The interface unit is used to perform interface security management on different functional interfaces in the edge system, and the different functional interfaces include at least a data upload interface, a model update interface, a control command interface, and a status monitoring interface.

[0018] In some embodiments, the edge model deployment and reasoning module includes:

[0019] A model deployment module, used to deploy the small and efficient model to various edge devices in clinical sites;

[0020] A data acquisition module, which is used to collect the patient's vital signs data and surgical process data obtained on the edge device in real time;

[0021] A risk warning module, used to input the collected data into the small and efficient model for performing different types of risk assessment and warning on the patient's clinical status;

[0022] The result output module is used to display and output risk assessment and warning results.

[0023] In some embodiments, the model update and synchronization mechanism module includes:

[0024] A cloud model updating module, when new clinical data is obtained, retrains and optimizes the complex model based on the new clinical data, and updates the small and efficient model based on the optimized complex model;

[0025] A first model synchronization module automatically synchronizes and updates the small and efficient models in the cloud system and the edge system;

[0026] An incremental update module, used to transmit the parameter difference between the small and efficient model in the cloud system and the small and efficient model in the edge system in an incremental update manner, so as to assist in realizing the synchronous update of the model;

[0027] The version control module is used to control and manage the version of the small and efficient model to ensure the traceability and rollback capability of the model update.

[0028] In some embodiments, the data flow and communication protocol module includes:

[0029] A data upload module is used to regularly upload the collected clinical data of patients to the cloud system through HTTP requests, so that the cloud system can complete the update of complex models and small and efficient models based on the new clinical data;

[0030] A second model synchronization module is used to obtain the updated small and efficient model of the cloud system and push it to each of the edge devices through an HTTP request;

[0031] A control command receiving module, used to receive the control command sent by the cloud system through HTTP request and push it to the edge device;

[0032] The status monitoring module is used to regularly report the status of the edge device and the operating status of the small and efficient model to the cloud system through HTTP requests, so that the cloud system can centrally monitor and manage the edge device and the small and efficient model.

[0033] In some embodiments, the interface unit further comprises:

[0034] A data management module interface, used to receive clinical data of the patient uploaded by the edge device;

[0035] Data query interface, used to query and obtain historical clinical data of a specified patient or a specified time period;

[0036] Data statistics interface, used to obtain statistical information and indicators of data;

[0037] A model management module interface, used to upload a new version of the small and efficient model to the cloud system;

[0038] Model query interface, used to query and obtain the list of available model versions;

[0039] Model download interface, used to push new model versions to edge devices;

[0040] Model update interface, used to trigger model update and distillation process;

[0041] Edge device management module interface, used to register new edge devices;

[0042] Device query interface, used to query and obtain information about registered devices;

[0043] Device configuration interface, used to modify the configuration parameters of edge devices;

[0044] A device control interface, used to send control commands to the edge device;

[0045] The reasoning and warning module interface is used to receive the risk assessment and warning results reported by the edge device;

[0046] Warning query interface, used to query and obtain warning records for a specified time period;

[0047] Early warning statistics interface, used to obtain early warning statistics and indicators;

[0048] A monitoring and diagnostic module interface, used to obtain the current status and system indicators of the edge device;

[0049] Log query interface, used to query and obtain logs on cloud systems and edge systems;

[0050] Health check interface, used to check the health status of cloud systems and edge systems;

[0051] Security and authentication module interface, used to authenticate users and obtain access tokens;

[0052] Authorization interface, used to manage user permissions and roles;

[0053] Key management interface, used to manage API keys and certificates.

[0054] In some embodiments, the knowledge distillation module obtains prediction results on the training data based on training complex models;

[0055] Use the soft labels of complex models as additional supervisory signals together with the original labels to train efficient models;

[0056] In the loss function of the efficient model, in addition to the original supervised loss, complex model knowledge that needs to be extracted is also introduced, including:

[0057] Label knowledge, the prediction results of complex models on data;

[0058] Intermediate layer output, the output of the intermediate layer of the complex model;

[0059] Gradient information, which is the gradient information generated by complex models during the training process;

[0060] By optimizing a comprehensive loss function, the efficient model fits the original labels and also learns the predictive behavior of the complex model.

[0061] In some embodiments, the large neural network model is a multi-layer perceptron mathematical model, which includes an input layer, an output layer, and multiple hidden layers, wherein the input layer X∈R m×n ,

[0062] m is the number of samples, n is the number of features;

[0063] The hidden layer contains h neurons, whose weights and biases are represented by W h ∈R n×h 、b h ∈R 1×h ;

[0064] There are q output label values, and the weight and bias parameters of the output layer are Wo∈R h×q 、b o ∈R 1×q ;

[0065] The output of the hidden layer and the output of the output layer are calculated by the following formulas:

[0066] In some embodiments 10, the risk warning system according to claim 2 is characterized in that when the knowledge distillation module generates a small and efficient model based on the complex model, for each clinical data, the original logical value generated by the complex model is recorded as v i , the logical value generated by the new model is z i , then the logical value produced by the small and efficient model is

[0067] The beneficial effects of the embodiments of the present invention include guiding the training of a small and efficient model by using the prediction results of a complex model as additional knowledge, so that the small and efficient model can not only learn the original training data, but also learn the prediction ability of the complex model. In addition, by performing model compression and knowledge transfer of complex models through knowledge distillation methods, the knowledge contained in the large complex model (teacher model) can be effectively transferred to the small and efficient model (student model), so that the small and efficient model can obtain a smaller model size and higher reasoning efficiency while maintaining high accuracy.

[0068] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings.

[0069] The technical solution of the present application is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 Schematic diagram of the structure of the risk prediction system in an embodiment of the present invention.

[0072] Figure 2 Schematic diagram of the structure of a risk prediction system in another embodiment of the present invention. DETAILED DESCRIPTION

[0073] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, but are not intended to limit the present invention.

[0074] It should be understood that various modifications may be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope of the present disclosure will occur to those skilled in the art.

[0075] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.

[0076] These and other characteristics of the invention will become apparent from the following description of a preferred form of embodiment given as a non-limiting example, with reference to the accompanying drawings.

[0077] It should also be understood that, although the invention has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the invention that have the characteristics recited in the claims and are therefore within the scope of protection defined thereby.

[0078] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0079] Specific embodiments of the present disclosure are described hereinafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of ways with substantially any suitable detailed structure.

[0080] This specification may use the phrases "in one embodiment," "in another embodiment," "in a further embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present disclosure.

[0081] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0082] like Figure 1 As shown, an embodiment of the present invention provides a risk warning system, including a cloud system and an edge system, or also referred to as a cloud and an edge end, wherein the cloud system is used to collect and process clinical data of different patients, and train a large neural network model based on the collected and processed data to obtain a complex model that can evaluate different clinical manifestations of patients and provide risk warnings, and then use knowledge distillation technology to distill the complex model into a small and efficient model;

[0083] The edge system is used to deploy the small and efficient model on various edge devices to utilize the small and efficient model to achieve clinical risk monitoring and early warning. The edge system is also used to synchronize data of the complex model and the small and efficient model.

[0084] Specifically, Figure 2 As shown, the cloud system in this embodiment includes:

[0085] The data collection and preprocessing module is used to obtain vital sign data, electronic medical record data, and expert annotation data of different patients in different clinical scenarios, and clean and standardize the obtained data;

[0086] A model training module, used to train a large neural network model using preprocessed data to obtain a complex model, wherein the complex model is used to perform patient emergency classification, vital sign risk warning, intraoperative hypotension and hypoxemia risk warning;

[0087] A model optimization module, used for optimizing the complex model;

[0088] The knowledge distillation module is used to transfer the knowledge of the optimized complex model through the knowledge distillation technology to obtain a small and efficient model.

[0089] The edge system includes:

[0090] An edge model deployment and reasoning module, which is used to deploy the small and efficient model to various edge devices at the clinical site to monitor and warn patients of risks based on the clinical data in the edge devices;

[0091] A model update and synchronization mechanism module, used to update the models on the cloud system and edge system using an incremental update method;

[0092] The data flow and communication protocol module is used to implement data exchange between various models in the cloud system and the edge system using the HTTP communication protocol;

[0093] The interface unit is used to perform interface security management on different functional interfaces in the edge system, and the different functional interfaces include at least a data upload interface, a model update interface, a control command interface, and a status monitoring interface.

[0094] Furthermore, in one embodiment, the edge model deployment and reasoning module includes:

[0095] A model deployment module, used to deploy the small and efficient model to various edge devices in clinical sites;

[0096] A data acquisition module, which is used to collect the patient's vital signs data and surgical process data obtained on the edge device in real time;

[0097] A risk warning module is used to input the collected data into the small and efficient model to perform different types of risk assessment and warning on the patient's clinical status; for example, to assess and warn the patient's emergency condition, abnormal vital signs, and intraoperative risks (such as hypotension, hypoxemia, etc.);

[0098] The result output module is used to display and output risk assessment and early warning results, for example, to present them to clinical medical staff to provide support for their decision-making.

[0099] In another embodiment, the model update and synchronization mechanism module includes:

[0100] A cloud model updating module, when new clinical data is obtained, retrains and optimizes the complex model based on the new clinical data, and updates the small and efficient model based on the optimized complex model;

[0101] A first model synchronization module automatically synchronizes and updates the small and efficient models in the cloud system and the edge system;

[0102] An incremental update module is used to transmit the difference in parameters between the small and efficient model in the cloud system and the small and efficient model in the edge system in an incremental update manner, so as to assist in realizing the synchronous update of the model; the setting of this module can reduce bandwidth and time overhead, and only the difference in model parameters is transmitted, not the entire model file;

[0103] The version control module is used to control and manage the version of the small and efficient model to ensure the traceability and rollback capability of the model update.

[0104] The data flow and communication protocol module includes:

[0105] A data upload module is used to regularly upload the collected clinical data of patients to the cloud system through HTTP requests, so that the cloud system can complete the update of complex models and small and efficient models based on the new clinical data;

[0106] A second model synchronization module is used to obtain the updated small and efficient model of the cloud system and push it to each of the edge devices through an HTTP request;

[0107] A control command receiving module is used to receive control commands sent by the cloud system via HTTP requests and push them to the edge device, such as starting / stopping inference, modifying configuration parameters, requesting diagnostic information, etc.;

[0108] The status monitoring module is used to regularly report the status of the edge device and the operating status of the small and efficient model to the cloud system through HTTP requests, so that the cloud system can centrally monitor and manage the edge device and the small and efficient model.

[0109] Furthermore, the interface unit also includes the following interfaces, that is, the risk warning system in this embodiment includes the following interfaces:

[0110] A data management module interface, used to receive clinical data of the patient uploaded by the edge device;

[0111] Data query interface, used to query and obtain historical clinical data of a specified patient or a specified time period;

[0112] Data statistics interface, used to obtain statistical information and indicators of data;

[0113] A model management module interface, used to upload a new version of the small and efficient model to the cloud system;

[0114] Model query interface, used to query and obtain the list of available model versions;

[0115] Model download interface, used to push new model versions to edge devices;

[0116] Model update interface, used to trigger model update and distillation process;

[0117] Edge device management module interface, used to register new edge devices;

[0118] Device query interface, used to query and obtain information about registered devices;

[0119] Device configuration interface, used to modify the configuration parameters of edge devices;

[0120] A device control interface, used to send control commands to the edge device, such as starting / stopping inference;

[0121] The reasoning and warning module interface is used to receive the risk assessment and warning results reported by the edge device;

[0122] Warning query interface, used to query and obtain warning records for a specified time period;

[0123] Early warning statistics interface, used to obtain early warning statistics and indicators;

[0124] A monitoring and diagnostic module interface, used to obtain the current status and system indicators of the edge device;

[0125] Log query interface, used to query and obtain logs on cloud systems and edge systems;

[0126] Health check interface, used to check the health status of cloud systems and edge systems;

[0127] Security and authentication module interface, used to authenticate users and obtain access tokens;

[0128] Authorization interface, used to manage user permissions and roles;

[0129] Key management interface, used to manage API keys and certificates.

[0130] In another embodiment, the knowledge distillation module obtains prediction results on the training data based on training a complex model, and uses the prediction structure of the complex model as a soft label to form an additional supervisory signal for training, and uses the supervisory signal together with the original label to train a small and efficient model;

[0131] In the loss function of the efficient model, in addition to the original supervised loss, complex model knowledge that needs to be extracted is also introduced, including:

[0132] Label knowledge, the prediction results of complex models on data, such as classification labels, regression values, etc.

[0133] Intermediate layer output, i.e. output data of the intermediate layer of the complex model;

[0134] Gradient information, which is the gradient information generated by complex models during the training process;

[0135] The knowledge distillation module optimizes the comprehensive loss function so that a small and efficient model can learn the predictive ability of a complex model while fitting the original labels.

[0136] The large neural network model is a multi-layer perceptron mathematical model, which includes an input layer, an output layer and multiple hidden layers, wherein the input layer X∈R m×n ,

[0137] m is the number of samples, n is the number of features;

[0138] The hidden layer contains h neurons, whose weights and biases are represented by W and h ∈R n×h , b h ∈R 1×h ;

[0139] There are q output label values, and the weight and bias parameters of the output layer are Wo∈R h×q , b o ∈R 1 ×q ;

[0140] The output of the hidden layer and the output of the output layer are calculated by the following formulas:

[0141]

[0142] Furthermore, when the knowledge distillation module generates a small and efficient model based on the complex model, for each clinical data, the original logical value generated by the complex model is recorded as v i, the logical value generated by the new model is z i , then the logical value produced by the small and efficient model is

[0143] Based on the contents disclosed in the above embodiments, the beneficial effects of this embodiment include guiding the training of a small and efficient model by using the prediction results of a complex model as additional knowledge, so that the small and efficient model can not only learn the original training data, but also learn the prediction ability of the complex model. In addition, by performing model compression and knowledge transfer of complex models through the knowledge distillation method, the knowledge contained in the large complex model (teacher model) can be effectively transferred to the small and efficient model (student model), so that the small and efficient model can obtain a smaller model size and higher reasoning efficiency while maintaining high accuracy.

[0144] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0145] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A risk early warning system, characterized in that: It includes a cloud system and an edge system. The cloud system is used to collect and process clinical data of different patients, and train a large neural network model based on the collected and processed data to obtain a complex model that can evaluate different clinical manifestations of patients and warn of risks. Then, the complex model is distilled into a small and efficient model using knowledge distillation technology; The edge system is used to deploy the small and efficient model on various edge devices to utilize the small and efficient model to achieve clinical risk monitoring and early warning. The edge system is also used to synchronize data of the complex model and the small and efficient model.

2. The risk early warning system according to claim 1, characterized in that: The cloud system includes: The data collection and preprocessing module is used to obtain vital sign data, electronic medical record data, and expert annotation data of different patients in different clinical scenarios, and clean and standardize the obtained data; A model training module, used to train a large neural network model using preprocessed data to obtain a complex model, wherein the complex model is used to perform patient emergency classification, vital sign risk warning, intraoperative hypotension and hypoxemia risk warning; A model optimization module, used for optimizing the complex model; The knowledge distillation module is used to transfer the knowledge of the optimized complex model through knowledge distillation technology to obtain a small and efficient model.

3. The risk early warning system according to claim 1, characterized in that: The edge system includes: An edge model deployment and reasoning module, which is used to deploy the small and efficient model to various edge devices at the clinical site to monitor and warn patients of risks based on the clinical data in the edge devices; A model update and synchronization mechanism module, used to update the models on the cloud system and edge system using an incremental update method; The data flow and communication protocol module is used to implement data exchange between various models in the cloud system and the edge system using the HTTP communication protocol; The interface unit is used to perform interface security management on different functional interfaces in the edge system, and the different functional interfaces include at least a data upload interface, a model update interface, a control command interface, and a status monitoring interface.

4. The risk early warning system according to claim 3 is characterized in that: The edge model deployment and reasoning module includes: A model deployment module, used to deploy the small and efficient model to various edge devices in clinical sites; A data acquisition module, which is used to collect the patient's vital signs data and surgical process data obtained on the edge device in real time; A risk warning module, used to input the collected data into the small and efficient model for performing different types of risk assessment and warning on the patient's clinical status; The result output module is used to display and output risk assessment and warning results.

5. The risk early warning system according to claim 3, characterized in that: The model update and synchronization mechanism module includes: A cloud model updating module, when new clinical data is obtained, retrains and optimizes the complex model based on the new clinical data, and updates the small and efficient model based on the optimized complex model; A first model synchronization module automatically synchronizes and updates the small and efficient models in the cloud system and the edge system; An incremental update module, used to transmit the parameter difference between the small and efficient model in the cloud system and the small and efficient model in the edge system in an incremental update manner, so as to assist in realizing the synchronous update of the model; The version control module is used to control and manage the version of the small and efficient model to ensure the traceability and rollback capability of the model update.

6. The risk early warning system according to claim 3, characterized in that: The data flow and communication protocol module includes: A data upload module is used to regularly upload the collected clinical data of patients to the cloud system through HTTP requests, so that the cloud system can complete the update of complex models and small and efficient models based on the new clinical data; A second model synchronization module is used to obtain the updated small and efficient model of the cloud system and push it to each of the edge devices through an HTTP request; A control command receiving module, used to receive the control command sent by the cloud system through HTTP request and push it to the edge device; The status monitoring module is used to regularly report the status of the edge device and the operating status of the small and efficient model to the cloud system through HTTP requests, so that the cloud system can centrally monitor and manage the edge device and the small and efficient model.

7. The risk early warning system according to claim 3, characterized in that: The interface unit also includes: A data management module interface, used to receive clinical data of the patient uploaded by the edge device; Data query interface, used to query and obtain historical clinical data of a specified patient or a specified time period; Data statistics interface, used to obtain statistical information and indicators of data; A model management module interface, used to upload a new version of the small and efficient model to the cloud system; Model query interface, used to query and obtain the list of available model versions; Model download interface, used to push new model versions to edge devices; Model update interface, used to trigger model update and distillation process; Edge device management module interface, used to register new edge devices; Device query interface, used to query and obtain information about registered devices; Device configuration interface, used to modify the configuration parameters of edge devices; A device control interface, used to send control commands to the edge device; The reasoning and warning module interface is used to receive the risk assessment and warning results reported by the edge device; Warning query interface, used to query and obtain warning records for a specified time period; Early warning statistics interface, used to obtain early warning statistics and indicators; A monitoring and diagnostic module interface, used to obtain the current status and system indicators of the edge device; Log query interface, used to query and obtain logs on cloud systems and edge systems; Health check interface, used to check the health status of cloud systems and edge systems; Security and authentication module interface, used to authenticate users and obtain access tokens; Authorization interface, used to manage user permissions and roles; Key management interface, used to manage API keys and certificates.

8. The risk early warning system according to claim 2, characterized in that: The knowledge distillation module obtains prediction results on the training data based on training complex models; Use the soft labels of complex models as additional supervisory signals together with the original labels to train efficient models; In the loss function of the efficient model, in addition to the original supervised loss, complex model knowledge that needs to be extracted is also introduced, including: Label knowledge, the prediction results of complex models on data; Intermediate layer output, the output of the intermediate layer of the complex model; Gradient information, which is the gradient information generated by complex models during the training process; By optimizing a comprehensive loss function, the efficient model fits the original labels and also learns the predictive behavior of the complex model.

9. The risk early warning system according to claim 2, characterized in that: The large neural network model is a multi-layer perceptron mathematical model, which includes an input layer, an output layer and multiple hidden layers, wherein the input layer X∈R m×n , m is the number of samples, n is the number of features; The hidden layer contains h neurons, whose weights and biases are represented by W h ∈R n×h , b h ∈R 1×h ; There are q output label values, and the weight and bias parameters of the output layer are Wo∈R h×q 、b o ∈R 1×q ; The output of the hidden layer and the output of the output layer are calculated by the following formulas:

10. The risk early warning system according to claim 2, characterized in that: When the knowledge distillation module generates a small and efficient model based on the complex model, for each clinical data, the original logical value generated by the complex model is recorded as v i , the logical value generated by the new model is z i , then the logical value produced by the small and efficient model is

Images