Automatic control methods, devices, computer equipment, readable storage media, and program products for ventilators

Abstract

This application relates to an automatic control method, apparatus, computer device, computer-readable storage medium, and computer program product for a ventilator. The method includes: acquiring status data and pressure control operations associated with each of multiple users in each respiratory cycle; for each respiratory cycle, configuring a control feedback component for the pressure control operation associated with that respiratory cycle based on the status data associated with the next respiratory cycle; using the status data associated with the respiratory cycle, the pressure control operation, the control feedback component, and the status data of the next respiratory cycle as sample elements to construct a training sample associated with that respiratory cycle, thereby training a target model. The target model is used to output a target pressure control operation based on the input status data associated with the respiratory cycle, so that after the ventilator operates according to the target pressure control operation, it can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, ensuring the user's respiratory state is normal in the next respiratory cycle.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an automatic control method, apparatus, computer equipment, computer-readable storage medium, and computer program product for a ventilator. Background Technology

[0002] A ventilator is an important medical device used to assist users in maintaining normal respiratory function. Traditional technology adjusts ventilator pressure by setting a target tidal volume. However, this control method can lead to excessively high pressure during automatic adjustment, causing user discomfort, affecting ventilator effectiveness, and potentially resulting in subsequent respiratory abnormalities. Traditional technology also detects breathing abnormalities and controls the ventilator to compensate for pressure, but this pressure control method has a delay and cannot adjust the pressure in time to prevent respiratory abnormalities from occurring.

[0003] Therefore, traditional automatic control methods for ventilators cannot adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, which makes it impossible to ensure the effectiveness of ventilator operation. Summary of the Invention

[0004] Therefore, it is necessary to provide an automatic control method, device, computer equipment, computer-readable storage medium, and computer program product for ventilators that can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, thereby ensuring the effectiveness of ventilator operation, in order to address the aforementioned technical problems.

[0005] In a first aspect, this application provides an automatic control method for a ventilator, comprising:

[0006] Acquire historical respiratory cycle data; historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0007] For each respiratory cycle, a control feedback segment is configured for the pressure control operation associated with the next respiratory cycle based on the status data associated with the next respiratory cycle; the control feedback segment represents the respiratory state of the user in the next respiratory cycle.

[0008] The state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle.

[0009] Training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0010] In one embodiment, acquiring historical respiratory cycle data includes:

[0011] Acquire status data and pressure control data generated by multiple users during each data acquisition period when using the ventilator;

[0012] For each user, the respiratory cycle is labeled for the status data and pressure control data of the target user in each data collection period;

[0013] The respiratory cycle status data and pressure control data of the target users that have been labeled in the same data collection period are combined into a data queue.

[0014] The collected data queue yields historical respiratory cycle data.

[0015] In one embodiment, the respiratory cycle history data includes multiple data queues, each data queue including: status data and pressure control data of a user corresponding to a labeled respiratory cycle during a data collection period;

[0016] For each respiratory cycle, based on the state data associated with the next respiratory cycle, a control feedback component is configured for the pressure control operation associated with the corresponding respiratory cycle, including:

[0017] Extract the target data queue from historical respiratory cycle data;

[0018] For each respiratory cycle marked in the target data queue, a control feedback component is configured for the pressure control operation represented by the pressure control data associated with the next respiratory cycle, based on the respiratory state represented by the state data associated with the next respiratory cycle.

[0019] In one embodiment, extracting a target data queue from historical respiratory cycle data includes:

[0020] Change the order of multiple data queues in the respiratory cycle history data;

[0021] Following the revised order, the data queues in the historical respiratory cycle data will be used as target data queues in sequence.

[0022] In one embodiment, a target model is trained based on multiple training samples, including:

[0023] Obtain the initial model and extract multiple target training samples from multiple training samples;

[0024] Based on multiple target training samples, the model parameters of the initial model are iteratively optimized until the model iteration optimization stopping condition is met, thus obtaining the target model.

[0025] In one embodiment, based on multiple target training samples, the model parameters of the initial model are iteratively optimized until the model iteration optimization stopping condition is met to obtain the target model, including:

[0026] Based on multiple target training samples, the model parameters of the initial model are iteratively optimized until the model iteration optimization stopping condition is met, resulting in at least one trained model.

[0027] Obtain a validation set for model evaluation, evaluate at least one trained model based on the validation set, and select at least one test model from at least one trained model based on the model evaluation results.

[0028] Obtain a test set for model testing, perform model testing on at least one test model based on the test set, and determine the target model from at least one test model based on the model testing results.

[0029] Secondly, this application also provides an automatic control device for a ventilator, comprising:

[0030] The respiratory cycle history data acquisition module is used to acquire respiratory cycle history data. The respiratory cycle history data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0031] The control feedback configuration module is used to configure control feedback for each respiratory cycle based on the status data associated with the next respiratory cycle. The control feedback represents the respiratory state of the user in the next respiratory cycle.

[0032] The training sample construction module is used to construct training samples associated with the respiratory cycle by taking the state data, pressure control operation, control feedback, and state data of the next respiratory cycle associated with the target respiratory cycle as sample elements.

[0033] The target model acquisition module is used to collect training samples from multiple users in each respiratory cycle to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user using the ventilator will have a normal breathing state in the next respiratory cycle.

[0034] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0035] Acquire historical respiratory cycle data; historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0036] For each respiratory cycle, a control feedback segment is configured for the pressure control operation associated with the next respiratory cycle based on the status data associated with the next respiratory cycle; the control feedback segment represents the respiratory state of the user in the next respiratory cycle.

[0037] The state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle.

[0038] Training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0039] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0040] Acquire historical respiratory cycle data; historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0041] For each respiratory cycle, a control feedback segment is configured for the pressure control operation associated with the next respiratory cycle based on the status data associated with the next respiratory cycle; the control feedback segment represents the respiratory state of the user in the next respiratory cycle.

[0042] The state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle.

[0043] Training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0045] Acquire historical respiratory cycle data; historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0046] For each respiratory cycle, a control feedback segment is configured for the pressure control operation associated with the next respiratory cycle based on the status data associated with the next respiratory cycle; the control feedback segment represents the respiratory state of the user in the next respiratory cycle.

[0047] The state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle.

[0048] Training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0049] The aforementioned automatic control method, apparatus, computer equipment, computer-readable storage medium, and computer program product for ventilators first acquire historical respiratory cycle data. This historical data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users are using the ventilator. Specifically, for each respiratory cycle, based on the status data associated with the next respiratory cycle, a control feedback score is configured for the pressure control operation associated with that respiratory cycle. This control feedback score represents the user's respiratory state in the next respiratory cycle. The status data associated with the respiratory cycle, the pressure control operation, the control feedback score, and the status data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle. This allows the model to learn how to output matching pressure control operations based on the input status data associated with the respiratory cycle, ensuring that the next respiratory cycle receives a control feedback score representing a normal respiratory state. In other words, the model learns how to ensure the user's respiratory state is normal in the next respiratory cycle. Based on this, the training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples, which are then used to train the target model. In this application, the target model can be used to output target pressure control operations based on the input respiratory cycle-related state data. This ensures that after the ventilator operates according to the target pressure control operations, the user's respiratory status is normal in the next respiratory cycle. Therefore, based on the trained target model, this application can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, ensuring the user's respiratory status is normal in the next respiratory cycle and guaranteeing the ventilator's operational effectiveness. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is an application environment diagram of the automatic control method for a ventilator in one embodiment;

[0052] Figure 2 This is a flowchart illustrating an automatic control method for a ventilator in one embodiment;

[0053] Figure 3 This is a flowchart illustrating the process of acquiring historical respiratory cycle data in one embodiment.

[0054] Figure 4This is a schematic diagram of the process of obtaining a training set by dividing the dataset after obtaining multiple data queues in one embodiment;

[0055] Figure 5 This is a schematic diagram of the process of constructing training samples and storing them in an experience pool in one embodiment, and then training the model based on the training samples in the experience pool.

[0056] Figure 6 This is a flowchart illustrating the process of training a target model and using the target model to achieve automatic control of a ventilator in one embodiment.

[0057] Figure 7 This is a structural block diagram of the automatic control device for a ventilator in one embodiment;

[0058] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] The automatic control method for ventilators provided in this application can be applied to, for example... Figure 1In the application environment shown, each ventilator 100 is equipped with a controller 102, a flow sensor 104 for collecting respiratory airflow data, and a pressure sensor 106 for collecting ventilator pressure data. The flow sensor 104 and pressure sensor 106 can transmit their collected data to the controller 102 in real time. The controller 102 can receive, process, and analyze the data in real time, and control the operation of the ventilator 100 based on the data analysis results, such as controlling the ventilator pressure. The controller 102 can integrate various control algorithms / data processing models, enabling intelligent operation and precise control of the ventilator 100. A server 200 can communicate with the controller 102 of the ventilator 100. The server 200 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. During model training, controller 102 can transmit data received and stored from flow sensor 104 and pressure sensor 106 during historical periods, as well as pressure control operations synchronously performed during ventilator operation during historical periods, to server 200. This allows server 200 to assist in model training based on the received historical data. After server 200 trains the target model, it can integrate the target model into the controller 102 of each ventilator 100. This allows each ventilator 100 to adjust the ventilator pressure promptly and accurately before respiratory abnormalities occur, based on the target model integrated into the controller 102. This ensures that the user of ventilator 100 maintains normal breathing in the next respiratory cycle, thus guaranteeing the effective operation of ventilator 100.

[0061] Specifically, in Figure 1 In the provided application environment, server 200 can communicate with the controller 102 of (one or more) ventilators 100 to obtain data associated with the operation of (one or more) ventilators 100 during historical periods (including respiratory airflow data, ventilator pressure data, and pressure control data). Server 200 can process the historical data, such as by constructing a training set, to obtain historical respiratory cycle data. The historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator (the pressure control operations can be determined by the pressure control data associated with the respiratory cycle).

[0062] Furthermore, in Figure 1In the provided application environment, for each respiratory cycle, server 200 can configure control feedback points (which represent the user's respiratory state in the next respiratory cycle) for the pressure control operation associated with the respiratory cycle according to the control feedback point configuration rules, based on the status data associated with the next respiratory cycle. Then, the status data associated with the respiratory cycle, the pressure control operation, the control feedback points, and the status data of the next respiratory cycle are used as sample elements to construct training samples associated with the respiratory cycle. This process aggregates training samples from multiple users in each respiratory cycle, resulting in multiple training samples. A target model is then trained based on these training samples and deployed to the controller 102 of each ventilator 100. The target model outputs target pressure control operations based on the input respiratory cycle-associated status data, ensuring that after the ventilator 100 with the target model operates according to the target pressure control operations, the user's respiratory state is normal in the next respiratory cycle. Based on this, when the ventilator 100 with the target model is running, it can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, based on the real-time collected status data associated with the current respiratory cycle. This ensures that the user of the ventilator 100 has a normal respiratory status in the next respiratory cycle, thus ensuring the operating effect of the ventilator 100.

[0063] It should be noted that the above-described process of training the target model can also be implemented without the aid of server 200. It can be implemented by the controller 102 in one ventilator 100 through data processing and analysis, or by the controllers 102 in multiple ventilators 100 jointly performing data processing and analysis. After the target model is trained, it can also be deployed on the controllers 102 of all ventilators 100.

[0064] In one exemplary embodiment, such as Figure 2 As shown, an automatic control method for a ventilator is provided, which can be applied to... Figure 1 Taking server 200 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0065] Step 202: Obtain historical respiratory cycle data; historical respiratory cycle data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0066] In this embodiment, a respiratory cycle can refer to a complete breathing process (including inspiration, expiration, and the pause between inspiration and expiration), or it can consist of multiple complete breathing processes (such as two or three complete breathing processes). The pressure control operations associated with each respiratory cycle can be determined by the pressure control data associated with each respiratory cycle. These pressure control operations include: increasing pressure, decreasing pressure, and maintaining pressure. The status data in this embodiment includes: respiratory airflow data and ventilator pressure data. Respiratory airflow data refers to parameters such as gas velocity and ventilation volume collected by the ventilator's flow sensor within the ventilator tubing when the user is using the ventilator. This data can be used to assess whether the user is experiencing hypoventilation or apnea. Ventilator pressure data refers to the airway pressure collected by the pressure sensor installed in the ventilator's airway tubing when the user is using the ventilator. Changes in airway pressure also help understand the user's breathing status, such as whether hypoventilation or apnea is occurring.

[0067] Optionally, the server can obtain data generated by multiple users using ventilators in a historical period by communicating with at least one controller. Furthermore, the server can perform data processing on the historical data, such as labeling the respiratory cycles of the historical data and selecting a portion to construct a training set, thereby obtaining historical respiratory cycle data.

[0068] Step 204: For each respiratory cycle, configure a control feedback segment for the pressure control operation associated with the next respiratory cycle based on the status data associated with the next respiratory cycle; the control feedback segment represents the respiratory state of the user in the next respiratory cycle.

[0069] Optionally, for each respiratory cycle, the server can configure control feedback points for the pressure control operation associated with the respiratory cycle according to the state data associated with the next respiratory cycle and the preset control feedback point configuration rules. This allows the model to learn how to output appropriate pressure control operations based on the real-time collected state data of the current respiratory cycle during subsequent model training, so that the user of the ventilator can maintain normal breathing in the next respiratory cycle.

[0070] For example, the control feedback sub-configuration rule may specifically include: (1) if the state data associated with the next respiratory cycle of the targeted respiratory cycle indicates that hypoventilation occurs, then configure control feedback sub-configuration for the pressure control operation associated with the targeted respiratory cycle. (2) If the state data associated with the next respiratory cycle of the targeted respiratory cycle indicates that apnea has occurred, then the pressure control operation associated with the targeted respiratory cycle is configured with control feedback. (3) If the status data associated with the next respiratory cycle of the targeted respiratory cycle indicates that there is no hypoventilation or apnea, then the pressure control operation associated with the targeted respiratory cycle is configured with control feedback. Based on this, during subsequent model training, the model can learn how to output appropriate pressure control operations based on the real-time collected state data of the current respiratory cycle, so that the pressure control operation can obtain control feedback. .

[0071] Step 206: Use the state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle as sample elements to construct training samples associated with the respiratory cycle.

[0072] Optionally, the server can target the respiratory cycle. Associated status data Pressure control operation Control feedback and the next breathing cycle Status data Each element is used as a sample element to construct the target respiratory cycle. Associated training samples Among them, status data Including: the respiratory cycle being targeted Associated respiratory airflow data (which can be a segment of ventilator airflow waveform) and ventilator pressure data (which can be a segment of ventilator pressure waveform); pressure control operation Including: in the targeted respiratory cycle Operations such as pressurization / depressurization / maintaining constant pressure are adopted; control feedback is provided. It is based on the next respiratory cycle Status data Configured for Any one of them.

[0073] Step 208: Collect training samples from multiple users in each respiratory cycle to obtain multiple training samples. Train the target model based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0074] Specifically, the target model is a trained reinforcement learning (RL) model. Reinforcement learning is a paradigm and methodology in machine learning that focuses on how an agent maximizes cumulative rewards through continuous trial and learning in an environment. The core of a reinforcement learning model is the interaction between the agent and the environment, optimizing the agent's behavioral strategy through a feedback loop of state, action, and reward. Based on this, during the model training process in this embodiment, the model can learn how to output appropriate pressure control operations (without pressure overshoot, improving the user's breathing) based on the state data of a certain breathing cycle. This allows the ventilator equipped with the target model to adjust the ventilator pressure in a timely and accurate manner before breathing abnormalities occur, ensuring that the user's breathing is normal in the next breathing cycle.

[0075] Optionally, the server can aggregate training samples from multiple users for each respiratory cycle, resulting in multiple training samples. The target model can then be trained based on these multiple training samples. After obtaining the target model, the server can deploy it to the controller of each ventilator. This allows ventilators with the target model deployed to adjust their pressure promptly and accurately based on real-time collected status data associated with the current respiratory cycle, before any respiratory abnormalities occur. This ensures the user's breathing is normal in the next respiratory cycle, thus guaranteeing the ventilator's effectiveness, rather than performing pressure compensation only after respiratory abnormalities occur.

[0076] The aforementioned automatic ventilator control method first acquires historical respiratory cycle data. This historical data includes the status data and pressure control operations associated with each user during each respiratory cycle when the ventilator is used. Specifically, for each respiratory cycle, a control feedback score is configured for the pressure control operation associated with the next respiratory cycle based on the status data associated with that cycle. This control feedback score represents the user's respiratory state in the next respiratory cycle. The status data associated with the respiratory cycle, the pressure control operation, the control feedback score, and the status data of the next respiratory cycle are used as sample elements to construct training samples associated with that respiratory cycle. This allows the model to learn how to output matching pressure control operations based on the input status data associated with the respiratory cycle, ensuring that the user receives a control feedback score representing a normal respiratory state in the next respiratory cycle. In other words, the model learns how to ensure the user's respiratory state is normal in the next respiratory cycle. Based on this, the training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples, which are then used to train the target model. In this application, the target model can be used to output target pressure control operations based on the input respiratory cycle-related state data. This ensures that after the ventilator operates according to the target pressure control operations, the user's respiratory status is normal in the next respiratory cycle. Therefore, based on the trained target model, this application can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, ensuring the user's respiratory status is normal in the next respiratory cycle and guaranteeing the ventilator's operational effectiveness.

[0077] In one exemplary embodiment, such as Figure 3 As shown, historical respiratory cycle data is obtained, including:

[0078] Step 302: Obtain the status data and pressure control data generated by each user when using the ventilator during each data acquisition period.

[0079] The data collection period can specifically refer to the time period during which data is collected within a historical period, such as several nights within that period. Status data includes respiratory airflow data and ventilator pressure data. Pressure control data can be used to determine pressure control operations, including: increasing pressure, decreasing pressure, and maintaining pressure. It should be noted that the data acquired in this embodiment is all collected under non-invasive assisted breathing (non-invasive respiratory therapy). Non-invasive assisted breathing specifically refers to: without establishing an artificial airway, primarily connecting the ventilator to the user through a mask (such as a nasal mask, face mask, or laryngeal mask), and providing positive pressure support through the ventilator to achieve assisted ventilation.

[0080] Optionally, the ventilator controller can collect and store status data and pressure control data generated by multiple users when using the ventilator in each data acquisition period, while the server can obtain the status data and pressure control data generated by each user when using the ventilator in each data acquisition period by communicating with the controller of at least one ventilator.

[0081] Step 304: For each user, label the respiratory cycle for the status data and pressure control data of the target user in each data collection period.

[0082] Furthermore, for each user, the server can label the status data and pressure control data of the target user in each data collection period according to a pre-determined respiratory cycle. For example, the status data and pressure control data generated by user 1 when using a ventilator at night 1 can be segmented according to the respiratory cycle and labeled with the respiratory cycle, so that the status data and pressure control data associated with user 1 in each respiratory cycle at night 1 can be determined later.

[0083] Step 306: Collect the respiratory cycle status data and pressure control data of the target user that have been labeled during the same data collection period into a data queue.

[0084] Optionally, the server can create a data queue containing the labeled respiratory cycle status data and pressure control data of the target user during the same data collection period. For example, the server can create a data queue containing the labeled status data and pressure control data for each user each night.

[0085] Step 308: Collect the obtained data queue to obtain historical respiratory cycle data.

[0086] Furthermore, the server can randomly select a portion of the collected data queues as historical respiratory cycle data, or it can take all the collected data queues as a large dataset, and then divide the large dataset according to a preset ratio to obtain a training set including multiple data queues, a validation set including multiple data queues, and a test set including multiple data queues. The training set can then be used as historical respiratory cycle data for subsequent model training.

[0087] For example, such as Figure 4 The diagram illustrates the process of obtaining a training set (historical respiratory cycle data) by partitioning the dataset after acquiring multiple data queues.

[0088] The server can obtain information on multiple users (such as...) during a historical period by communicating with the control of at least one ventilator. Figure 4The system generates status and pressure control data for each user (user 1...user n) each night while using the ventilator. The status data includes respiratory flow data and ventilator pressure data. Furthermore, for each user, the server can label the respiratory cycle on the nightly status and pressure control data. For example... Figure 4 As shown, respiratory cycles are labeled using the status data and pressure control data of object 1 at night 1, so that the status data and pressure control data associated with each respiratory cycle at night 1 can be determined later.

[0089] Furthermore, such as Figure 4 As shown, the server can divide the nightly labeled data of each user into separate data queues. Specifically, it can group the nightly labeled respiratory cycle status data and pressure control data of each user into one data queue. Based on this, the server can aggregate the resulting data queues to obtain a large dataset. Then, it can divide the large dataset according to a preset ratio to obtain a training set including multiple data queues, a test set including multiple data queues, and a validation set including multiple data queues. The training set can be used as historical respiratory cycle data for subsequent model training.

[0090] In this embodiment, by labeling the respiratory cycles of data generated by multiple users when using ventilators in historical periods, it is possible to subsequently determine the status data and pressure control operations associated with each respiratory cycle in each data queue. This facilitates the configuration of control feedback scores for the pressure control operations associated with each respiratory cycle, thereby making it easier to construct training samples.

[0091] In one embodiment, the respiratory cycle historical data includes multiple data queues, each containing: status data and pressure control data of a user corresponding to a labeled respiratory cycle during a data acquisition period. Based on this, for each respiratory cycle, a control feedback component is configured for the pressure control operation associated with the corresponding respiratory cycle, according to the status data associated with the next respiratory cycle, including:

[0092] Extract the target data queue from historical respiratory cycle data;

[0093] For each respiratory cycle marked in the target data queue, a control feedback component is configured for the pressure control operation represented by the pressure control data associated with the next respiratory cycle, based on the respiratory state represented by the state data associated with the next respiratory cycle.

[0094] Optionally, the server can sequentially extract data queues from the historical respiratory cycle data (training set) as target data queues according to the order of the data queues. Further, for each respiratory cycle marked in the target data queue, the server can configure control feedback for the pressure control operation represented by the pressure control data associated with the target respiratory cycle, based on the respiratory state represented by the state data associated with the next respiratory cycle.

[0095] For example, if the target data queue includes n respiratory cycles, the server can configure control feedback segments sequentially for the pressure control operations associated with the first n-1 respiratory cycles in the target data queue. For instance, based on the status data associated with the second respiratory cycle in the target data queue... This determines the respiratory state associated with the second respiratory cycle in the target data queue, thereby providing a basis for the pressure control operation characterized by the pressure control data associated with the first respiratory cycle in the target data queue. Configuration control feedback And associate the status data of the first respiratory cycle in the target data queue. Pressure control operation Control feedback Status data associated with the second respiratory cycle These elements are used as sample elements associated with the first respiratory cycle, thereby constructing the training samples associated with the first respiratory cycle in the target data queue. And so on, training samples associated with the 2nd to (n-1)th respiratory cycles in the target data queue are constructed sequentially. And in constructing the training samples associated with the (n-1)th respiratory cycle in the target data queue. Afterwards, the processing of the target data queue is considered complete, and the next data queue is selected as the new target data queue. Each time a training sample is obtained, the server can store it in the experience pool for use in subsequent model training.

[0096] It should be noted that, in addition to sequentially selecting data queues from the historical respiratory cycle data to configure control feedback scores, this application can also configure control feedback scores for data queues in the historical respiratory cycle data in a multi-threaded and parallel manner.

[0097] In this embodiment, by configuring a control feedback score for the pressure control operation associated with each respiratory cycle in each data queue, the model can learn how to output a matching pressure control operation based on the input respiratory cycle-associated state data when training the model based on training samples. This allows the control feedback score of the pressure control operation to be configured. That is, to ensure that the person using the ventilator has a normal respiratory status in the next respiratory cycle.

[0098] In some embodiments, extracting a target data queue from historical respiratory cycle data includes:

[0099] Change the order of multiple data queues in the respiratory cycle history data;

[0100] Following the revised order, the data queues in the historical respiratory cycle data will be used as target data queues in sequence.

[0101] Optionally, considering that the initial order of multiple data queues in the respiratory cycle history data may be: the data queue corresponding to object 1 at night 1, the data queue corresponding to object 1 at night 2, etc., in order to improve the model training effect, the server can randomly shuffle the order of multiple data queues in the respiratory cycle history data, and then use the data queues in the respiratory cycle history data as target data queues in sequence according to the changed order.

[0102] In this embodiment, the order of the data queues in the training set can be shuffled first, and then the data queues in the training set can be processed sequentially according to the shuffled order.

[0103] In one embodiment, a target model is trained based on multiple training samples, including:

[0104] Obtain the initial model and extract multiple target training samples from multiple training samples;

[0105] Based on multiple target training samples, the model parameters of the initial model are iteratively optimized until the model iteration optimization stopping condition is met, thus obtaining the target model.

[0106] The stopping condition for model iteration optimization can be flexibly configured according to the needs of the actual application scenario. For example, model training can be stopped when the loss function obtained after a certain training iteration is less than a preset threshold. The initial model can be a model built based on a reinforcement learning network. In this embodiment, multiple reinforcement learning networks can be used as initial models so that the best model can be selected later.

[0107] Optionally, for each initial model, the server can randomly select multiple target training samples from an experience pool containing multiple training samples, and sequentially input these target training samples into the model. After each input of a target training sample, the server calculates the loss function based on the model's output, and then performs gradient descent to optimize the model parameters. The next target training sample is then input into the optimized model, and this process is repeated until the model iteration stops, resulting in a trained model. Furthermore, the server can select the final target model from at least one trained model based on the validation and test sets.

[0108] More specifically, in this embodiment, the model parameters of the initial model are iteratively optimized based on multiple target training samples until the model iteration optimization stopping condition is met to obtain the target model. This includes: iteratively optimizing the model parameters of the initial model based on multiple target training samples until the model iteration optimization stopping condition is met to obtain at least one trained model; obtaining a validation set for model evaluation; evaluating the at least one trained model based on the validation set; and selecting at least one test model from the at least one trained model based on the model evaluation results; obtaining a test set for model testing; testing the at least one test model based on the test set; and determining the target model from the at least one test model based on the model test results. The validation set can be used to select the best model from the at least one trained model, and the test set is used to test the performance of the selected best model.

[0109] It should be noted that if the expected results are not obtained when using the validation set, the server can retrain the model; similarly, if the expected results are not obtained when using the test set, the server can also retrain the model to ensure that the final target model can output control feedback based on the state data associated with the input respiratory cycle. Pressure control operation, that is, ensuring that the user of the ventilator has a normal respiratory status in the next respiratory cycle.

[0110] For example, such as Figure 5 The diagram illustrates a process for constructing training samples and storing them in an experience pool, then training a model based on the training samples in the experience pool.

[0111] (1) Regarding the process of constructing training samples and storing them in the experience pool: First, determine the current data queue (target data queue) and the current respiratory cycle from the training set, and determine the state data and pressure control operations taken in that respiratory cycle. Further, based on the state data in the next respiratory cycle, configure control feedback points for the pressure control operations taken in that respiratory cycle. Then, based on the state data, pressure control operations, control feedback points, and state data in the next respiratory cycle, construct training samples for that respiratory cycle and store the constructed training samples in the experience pool. If the current data queue has not been processed, continue to construct training samples for the next respiratory cycle in that data queue; if the current data queue has been processed, select the next data queue from the training set for processing.

[0112] (2) Regarding the process of training the model based on training samples in the experience pool: Taking a certain reinforcement learning network as an example, the network parameters can be randomly initialized based on the reinforcement learning network to obtain reinforcement learning networks with different initialized network parameters. (weight is) and reinforcement learning networks (weight is) (Using a training sample of a specific target drawn from the experience pool). For example, the respiratory cycle state data associated with the target training sample can be used. and pressure control operation Input reinforcement learning network Reinforcement learning network Output value The state data of the next respiratory cycle And the pressure control operations that can be taken input reinforcement learning network Reinforcement learning network Output predicted value :

[0113]

[0114] in, is the discount factor, which is located between [0,1). This indicates the pressure control measures that can be taken.

[0115] Furthermore, the server can calculate the loss value according to a preset loss function rule. and predicted value Calculate the loss function after this training. Gradient descent is performed based on the calculation results of the loss function to update the reinforcement learning network. The updated model parameters are then copied to the reinforcement learning network. The loss function measures the difference between the model's predictions and the actual data, while gradient descent is an optimization algorithm that aims to minimize the loss function by updating the network parameters (model parameters) in this embodiment along the negative direction of the gradient (the direction in which the loss function decreases the fastest). In this embodiment, the process of selecting target training samples and inputting them into the network → calculating the loss function → optimizing the network parameters (model parameters) can be repeated until the model iteration optimization stopping condition is met, resulting in the trained model.

[0116] Furthermore, in this embodiment, various types of reinforcement learning networks can be employed to obtain multiple trained models for selection. For example, multiple trained models can be evaluated based on a validation set to assess model performance, and at least one test model with expected performance can be selected from the multiple trained models based on the evaluation results. Then, at least one test model can be tested based on a test set to test its generalization ability, and the final target model can be determined from at least one test model based on the test results.

[0117] In this embodiment, the model can be trained, screened, and tested using a training set, a validation set, and a test machine to obtain the final target model. This allows the ventilator deployed with the target model to determine the matching target pressure control operation based on the real-time collected state data of a certain respiratory cycle. After running the target pressure control operation, the user of the ventilator will have a normal breathing state in the next respiratory cycle.

[0118] In one specific embodiment, such as Figure 6 The diagram shows a flowchart illustrating the training of a target model and the implementation of automatic ventilator control using that model.

[0119] The specific steps involved in training a model on a server include:

[0120] Step 602: Obtain the status data and pressure control data generated by each user when using the ventilator during each data acquisition period, and label the respiratory cycle respectively;

[0121] Step 604: Collect the respiratory cycle status data and pressure control data of each user during the same data collection period into a data queue;

[0122] Step 606: Divide the aggregated data queues into training set, test set and validation set;

[0123] Step 608: Randomly shuffle the order of multiple data queues in the training set, and use the data queues in the training set as target data queues in sequence according to the shuffled order.

[0124] Step 610: For each respiratory cycle marked in the target data queue, configure a control feedback component for the pressure control operation represented by the pressure control data associated with the target respiratory cycle, based on the status data associated with the next respiratory cycle of the target respiratory cycle.

[0125] Step 612: Use the state data associated with the respiratory cycle, pressure control operation, control feedback, and state data of the next respiratory cycle as sample elements to construct training samples associated with the respiratory cycle.

[0126] Step 614: Store the constructed training samples into the experience pool, and extract multiple target training samples from the experience pool.

[0127] Step 616: Based on multiple target training samples, iteratively optimize the model parameters of the initial model until the model iteration optimization stopping condition is met, and obtain at least one trained model.

[0128] Step 618: Evaluate at least one trained model based on the validation set, and select at least one test model from at least one trained model according to the model evaluation results.

[0129] Step 620: Perform model testing on at least one test model based on the test set, and determine the target model from at least one test model based on the model test results;

[0130] Step 622: Deploy the target model on the ventilator;

[0131] In practical applications, a ventilator equipped with a target model can perform the following steps:

[0132] Step 624: Based on the target model and the state data collected in real time during a certain respiratory cycle, determine the matching target pressure control operation;

[0133] Step 626: Operate according to the target pressure control operation to ensure that the user of the ventilator has a normal breathing status in the next respiratory cycle.

[0134] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0135] Based on the same inventive concept, this application also provides an automatic ventilator control device for implementing the aforementioned automatic ventilator control method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the automatic ventilator control device provided below can be found in the limitations of the automatic ventilator control method described above, and will not be repeated here.

[0136] In one exemplary embodiment, such as Figure 7 As shown, an automatic control device for a ventilator is provided, including: a respiratory cycle history data acquisition module 702, a control feedback configuration module 704, a training sample construction module 706, and a target model acquisition module 708, wherein:

[0137] The respiratory cycle history data acquisition module is used to acquire respiratory cycle history data. The respiratory cycle history data includes the status data and pressure control operations associated with each user in each respiratory cycle when multiple users of the ventilator are using the ventilator.

[0138] The control feedback configuration module is used to configure control feedback for each respiratory cycle based on the status data associated with the next respiratory cycle. The control feedback represents the respiratory state of the user in the next respiratory cycle.

[0139] The training sample construction module is used to construct training samples associated with the respiratory cycle by taking the state data, pressure control operation, control feedback, and state data of the next respiratory cycle associated with the target respiratory cycle as sample elements.

[0140] The target model acquisition module is used to collect training samples from multiple users in each respiratory cycle to obtain multiple training samples. The target model is trained based on the multiple training samples. The target model is used to output target pressure control operations based on the state data associated with the input respiratory cycle, so that after the ventilator runs according to the target pressure control operations, the user using the ventilator will have a normal breathing state in the next respiratory cycle.

[0141] The aforementioned automatic control device for the ventilator first acquires historical respiratory cycle data. This historical data includes the status data and pressure control operations associated with each user during each respiratory cycle when the ventilator is used. Specifically, for each respiratory cycle, based on the status data associated with the next respiratory cycle, a control feedback score is configured for the pressure control operation associated with that respiratory cycle. This control feedback score represents the user's respiratory state in the next respiratory cycle. The status data associated with the respiratory cycle, the pressure control operation, the control feedback score, and the status data of the next respiratory cycle are used as sample elements to construct training samples associated with that respiratory cycle. This allows the model to learn how to output matching pressure control operations based on the input status data associated with the respiratory cycle, ensuring that the user receives a control feedback score representing a normal respiratory state in the next respiratory cycle. In other words, the model learns how to ensure the user's respiratory state is normal in the next respiratory cycle. Based on this, the training samples from multiple users in each respiratory cycle are collected to obtain multiple training samples, which are then used to train the target model. In this application, the target model can be used to output target pressure control operations based on the input respiratory cycle-related state data. This ensures that after the ventilator operates according to the target pressure control operations, the user's respiratory status is normal in the next respiratory cycle. Therefore, based on the trained target model, this application can adjust the ventilator pressure in a timely and accurate manner before respiratory abnormalities occur, ensuring the user's respiratory status is normal in the next respiratory cycle and guaranteeing the ventilator's operational effectiveness.

[0142] In one embodiment, the respiratory cycle historical data acquisition module further includes: a historical data acquisition unit, used to acquire status data and pressure control data generated by multiple users when using the ventilator in each data acquisition period; a respiratory cycle labeling unit, used to label the respiratory cycle for each user's status data and pressure control data in each data acquisition period; a data queue determination unit, used to collect the status data and pressure control data of the user whose respiratory cycles have been labeled in the same data acquisition period as a data queue; and a data queue aggregation unit, used to aggregate the obtained data queues to obtain respiratory cycle historical data.

[0143] In one embodiment, the respiratory cycle history data includes multiple data queues, each containing: status data and pressure control data of a user corresponding to a labeled respiratory cycle during a data acquisition period. Based on this, the control feedback configuration module further includes: a data queue extraction unit, used to extract a target data queue from the respiratory cycle history data; and a control feedback configuration unit, used to configure a control feedback score for each respiratory cycle labeled in the target data queue, based on the respiratory state represented by the status data associated with the next respiratory cycle of the targeted respiratory cycle, and for the pressure control operation represented by the pressure control data associated with the targeted respiratory cycle.

[0144] In one embodiment, the data queue extraction unit is specifically used to: change the order of multiple data queues in the respiratory cycle history data; and, according to the changed order, sequentially use the data queues in the respiratory cycle history data as target data queues.

[0145] In one embodiment, the target model acquisition module includes: an initial model acquisition unit, used to acquire an initial model and extract multiple target training samples from multiple training samples; and a model iteration optimization unit, used to iteratively optimize the model parameters of the initial model based on the multiple target training samples until the model iteration optimization stopping condition is met, thereby obtaining the target model.

[0146] In one embodiment, the model iterative optimization unit is specifically used to: iteratively optimize the model parameters of the initial model based on multiple target training samples until the model iterative optimization stopping condition is met, thereby obtaining at least one trained model; obtain a validation set for model evaluation, evaluate the at least one trained model based on the validation set, and select at least one test model from the at least one trained model according to the model evaluation results; obtain a test set for model testing, perform model testing on the at least one test model based on the test set, and determine the target model from the at least one test model according to the model testing results.

[0147] The modules in the aforementioned automatic control device for ventilators can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0148] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores automatic control data for the ventilator. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an automatic control method for the ventilator.

[0149] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0151] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0152] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0153] It should be noted that the user information (including but not limited to information about the ventilator used by the user, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0154] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0155] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0156] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A model training method for a ventilator, characterized in that, The method comprises: acquiring state data and pressure control data generated by each of a plurality of use objects when using a ventilator at each data collection period; for each of the use objects, labeling a breathing cycle for the state data and the pressure control data of the use object at each data collection period; using the state data and the pressure control data of the use object at the same data collection period as a data queue; collecting the data queue to obtain breathing cycle history data; the breathing cycle history data comprises a plurality of data queues, and each of the data queues comprises state data and pressure control data of a use object at a data collection period corresponding to a labeled breathing cycle; extracting a target data queue from the breathing cycle history data for each of the breathing cycles; for each breathing cycle labeled in the target data queue, configuring a control feedback score for a pressure control operation associated with the breathing cycle according to a breathing state represented by state data associated with a next breathing cycle of the breathing cycle; using the state data, the pressure control operation, the control feedback score, and the state data of the next breathing cycle as sample elements to construct a training sample associated with the breathing cycle; collecting training samples of each of the use objects at each breathing cycle to obtain a plurality of training samples, and training a target model based on the plurality of training samples.

2. The method of claim 1, wherein, The target data queue is extracted from the breathing cycle history data, comprising: changing the order of the plurality of data queues in the breathing cycle history data; according to the changed order, the data queues in the breathing cycle history data are sequentially used as target data queues.

3. The method of claim 1, wherein, The target model is trained based on the plurality of training samples, comprising: obtaining an initial model and extracting a plurality of target training samples from the plurality of training samples; based on the plurality of target training samples, iteratively optimizing model parameters of the initial model until a model iteration optimization stop condition is reached to obtain a target model.

4. The method of claim 3, wherein, Based on the plurality of target training samples, iteratively optimizing model parameters of the initial model until a model iteration optimization stop condition is reached to obtain a target model, comprising: based on the plurality of target training samples, iteratively optimizing model parameters of the initial model until a model iteration optimization stop condition is reached to obtain at least one trained model; obtaining a verification set for model evaluation, performing model evaluation on the at least one trained model based on the verification set, and selecting at least one test model from the at least one trained model according to the model evaluation result; obtaining a test set for model testing, performing model testing on the at least one test model based on the test set, and determining a target model from the at least one test model according to the model testing result.

5. A breathing machine automatic control device, characterized by, The device comprises: The respiratory cycle history data acquisition module is configured to acquire state data and pressure control data generated by each of a plurality of use objects when using a ventilator in each data collection period; for each of the use objects, label a respiratory cycle for the state data and the pressure control data of the use object in each data collection period; and take the state data and the pressure control data of the use object in the same data collection period and having the labeled respiratory cycle as a data queue. The respiratory cycle history data is obtained by collecting the data queues. The respiratory cycle history data includes a plurality of data queues, and each of the data queues includes state data and pressure control data of a use object in a data collection period and having a labeled respiratory cycle. The control feedback score configuration module is configured to extract a target data queue from the respiratory cycle history data; for each respiratory cycle labeled in the target data queue, configure a control feedback score for pressure control operation associated with the respiratory cycle according to a respiratory state represented by state data associated with a next respiratory cycle of the respiratory cycle; The training sample construction module is configured to construct a training sample associated with the respiratory cycle by taking the state data associated with the respiratory cycle, the pressure control operation, the control feedback score, and the state data of the next respiratory cycle as sample elements, respectively. The target model obtaining module is configured to obtain a plurality of training samples by collecting training samples of each of the plurality of use objects in each respiratory cycle, and train a target model according to the plurality of training samples. The target model is configured to output a target pressure control operation according to input state data associated with a respiratory cycle, so that the use object using the ventilator has a normal respiratory state in a next respiratory cycle after the ventilator operates according to the target pressure control operation.

6. The apparatus of claim 5, wherein, The control feedback score configuration module is further configured to change the order of the plurality of data queues in the respiratory cycle history data, and take the data queues in the respiratory cycle history data in the changed order as target data queues in sequence.

7. The apparatus of claim 5, wherein, The target model obtaining module is further configured to obtain an initial model, extract a plurality of target training samples from the plurality of training samples, and iteratively optimize model parameters of the initial model based on the plurality of target training samples until a model iterative optimization stop condition is reached to obtain the target model.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 4.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 4.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 4.

Images