Human factor intelligent model training and prediction method, edge computing device, system and medium

Abstract

The invention discloses a human factor intelligent model training and prediction method, an edge computing device, a system and a medium, and relates to the field of human factor data evaluation.The human factor intelligent evaluation model training method comprises the steps that a candidate training scheme of a human factor intelligent model is obtained, the candidate training scheme specifies a training data set and a to-be-trained model, and the training data set and the to-be-trained model are selected according to the candidate training scheme; the training data set is extracted from a plurality of human factor data sources according to training requirements; determining a target training scheme from the candidate training schemes based on a received selection instruction; and obtaining a corresponding training data set based on the target training scheme, and training the to-be-trained model based on the corresponding training data set to obtain a trained model which can be used for predicting the human factor state. According to the training method of the human factor intelligent model, various types of training schemes can be configured according to the training requirements of the user, the human factor state is predicted by utilizing the trained model, and the training method is more flexible.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and human factors intelligence, and in particular to a method for training and predicting human factors intelligence models, an edge computing device, a system, and a medium. Background Technology

[0002] Existing large-scale AI evaluation models are often trained on broad datasets, which makes it difficult to meet the application needs of specific vertical fields, and also fails to meet the functional requirements of human factors engineering in specific vertical fields, such as model training and state prediction. Summary of the Invention

[0003] Therefore, the purpose of this application is to propose a human factors intelligence model training and prediction method, edge computing device, system, medium and program product, which can configure various types of training schemes according to the user's training needs, and use the trained model to predict human factors status, which is more flexible and can design and train corresponding human factors intelligence models for a specific vertical field.

[0004] This application provides a method for training a human factors intelligence model. The method includes: obtaining candidate training schemes for the human factors intelligence model, wherein the candidate training schemes specify a training dataset and a model to be trained, wherein the training dataset is extracted from multiple human factors data sources according to training requirements; determining a target training scheme from the candidate training schemes based on a received selection instruction; obtaining a corresponding training dataset based on the target training scheme, and training the model to be trained based on the corresponding training dataset to obtain a trained model, wherein the trained model can be used to predict human factors states.

[0005] For example, before obtaining candidate training schemes for the human factors intelligence model, the method further includes: obtaining training scheme configuration information; and generating candidate training schemes based on the training scheme configuration information.

[0006] For example, the training scheme configuration information includes training data source configuration information, and the step of generating a candidate training scheme based on the training scheme configuration information includes: obtaining the training data source configuration information; obtaining corresponding data from the human-machine environment synchronization platform based on the training data source configuration information, processing the obtained data, and using the processed data as the training dataset for the candidate training scheme.

[0007] For example, generating a candidate training scheme based on the training scheme configuration information includes: obtaining the type information of the model to be trained; determining the algorithm model corresponding to the type information of the model to be trained from the candidate algorithm models based on the type information of the model to be trained, and using it as the model to be trained for the candidate training scheme.

[0008] For example, the acquired data includes human factors segments, and the processing of the acquired data includes: acquiring tag information, wherein the tag information is used to label the state of the human factors segments; and adding the tag information to the human factors segments.

[0009] For example, obtaining the corresponding data from the human-machine environment synchronization platform includes: reading a configuration file created in the system background to obtain the corresponding data from the human-machine environment synchronization platform, wherein the configuration file includes information for characterizing the data source.

[0010] For example, the training data source configuration information includes baseline data configuration information and non-baseline data configuration information. The step of obtaining corresponding data from the human-machine environment synchronization platform based on the training data source configuration information includes: obtaining baseline data from the human-machine environment synchronization platform based on the baseline data configuration information, and obtaining non-baseline data from the human-machine environment synchronization platform based on the non-baseline data configuration information. The training dataset includes the baseline data and the non-baseline data, and / or the baseline data is used to verify or optimize the performance of the human factors intelligence model.

[0011] For example, the information used to characterize the data source includes information about the acquisition device, which includes at least one of the following: a smart wearable sensor, a biosensor, a mobile portable sensor, an eye tracker; and / or,

[0012] The training data sources include at least one of the following: human-related data sources, machine-related data sources, human-computer interaction-related data sources, and environment-related data sources; wherein, the human-related data sources include one or a combination of the following: skin conductance and temperature data, pulse data, blood pressure data, blood oxygen data, electrocardiogram data, electromyography data, muscle oxygenation data, respiratory data, biomechanical data, near-infrared brain imaging data, electroencephalogram data, transcranial stimulation data, heart rate variability data, heart rate data, image / video data, sound data, eye-tracking data, and gesture or movement data; the machine-related data sources include one or a combination of the following: machine movement... The data sources include: line data, fault alarm data, machine control data, machine model data, machine communication data, and machine positioning data; the human-computer interaction related data sources include one or a combination of the following: human-computer voice interaction data, human-computer text interaction data, human-computer touch interaction data, human-computer gesture or action interaction data, human-computer EEG interaction data, human-computer eye-tracking interaction data, and human-computer facial expression interaction data; the environment related data sources include one or a combination of the following: location data, humidity data, temperature data, color data, brightness data, weather data, road condition data, traffic data, stimulus signal data, and event or signal tagging data; and / or,

[0013] Candidate algorithm models include at least one of the following: linear discriminant analysis model, random forest model, support vector machine algorithm model, logistic regression model, decision tree model, and classification algorithm model.

[0014] Another embodiment of this application provides a human factors intelligent prediction method, the method comprising: extracting a dataset to be predicted from multiple human factors data sources; determining a target prediction scheme from candidate prediction schemes based on a received prediction scheme selection operation, wherein the target prediction scheme specifies a trained model; and predicting the human factors state of the dataset to be predicted using the trained model based on the target prediction scheme.

[0015] For example, determining the target prediction scheme from the candidate prediction schemes based on the received prediction scheme selection operation includes: obtaining prediction scheme configuration information, wherein the prediction scheme configuration information includes target prediction scheme path information; and determining the target prediction scheme from the candidate prediction schemes based on the target prediction scheme path information.

[0016] For example, the dataset to be predicted includes at least one human factor segment, and the step of using the trained model to predict the human factor state of the dataset to be predicted includes: using the trained model to predict the dataset to be predicted, predicting the label of each human factor segment; scoring the human factor segment based on the label score corresponding to the predicted label to obtain a score for the human factor segment, and obtaining the human factor state prediction result based on the score of the human factor segment.

[0017] For example, the human factor segments include multiple segments, and the step of scoring each human factor segment based on the score corresponding to the tag, and obtaining the human factor status prediction result based on the human factor segment score, includes: determining a comprehensive score value based on the scores of the multiple human factor segments, wherein the comprehensive score value is obtained by weighting the tag scores corresponding to the multiple human factor segments; and obtaining the human factor status prediction result based on the warning level configuration information and the comprehensive score value.

[0018] For example, the human condition prediction result includes a warning level, and obtaining the human condition prediction result based on the warning level configuration information and the comprehensive score includes: determining the warning level based on the warning level configuration information when the comprehensive score meets the warning conditions.

[0019] Another embodiment of this application provides a human factors intelligence system, the human factors intelligence system including a human factors intelligence platform, the human factors intelligence platform including a training end and a prediction end; the training end is used to execute the above-described human factors intelligence model training method; the prediction end is used to execute the above-described human factors intelligence prediction method.

[0020] For example, the human factors intelligent system further includes a human-machine environment synchronization platform, which is connected to the human-machine environment synchronization platform. The human-machine environment synchronization platform is used to store, acquire, and / or process human factors data collected by human factors signal acquisition devices.

[0021] Another embodiment of this application provides an edge computing device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the training steps of the human factors intelligence model corresponding to the above method and / or the human factors intelligence prediction steps corresponding to the above method.

[0022] Another embodiment of this application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements the training steps of the human factors intelligence model corresponding to the above method and / or the human factors intelligence prediction steps corresponding to the above method.

[0023] Another embodiment of this application provides a computer program product, which includes instructions that, when executed by a processor of a computer device, enable the computer device to perform the steps of the method described in any of the above embodiments.

[0024] In the above embodiments, candidate training schemes for the human factors intelligent model are obtained. Each candidate training scheme specifies a training dataset and a model to be trained. The training dataset is extracted from multiple human factors data sources according to training requirements. Based on the received selection instruction, a target training scheme is determined from the candidate training schemes. A corresponding training dataset is obtained based on the target training scheme, and the model to be trained is trained based on the corresponding training dataset to obtain a trained model. The trained model can be used to evaluate human factors status. The human factors intelligent model training method of the present invention can configure various types of training schemes according to the user's training needs and use the trained model to predict human factors status, thus providing greater flexibility. Attached Figure Description

[0025] Figure 1 A flowchart illustrating the training method for the human-computer intelligence model provided in this application's embodiments;

[0026] Figure 2 A flowchart illustrating the configuration candidate training schemes provided in the embodiments of this application;

[0027] Figure 3 A flowchart for generating candidate training schemes based on training scheme configuration information provided in this application embodiment;

[0028] Figure 4 A schematic diagram of the training scheme configuration interface provided in the embodiments of this application;

[0029] Figure 5 A flowchart of the algorithm model for configuring candidate training schemes provided in the embodiments of this application;

[0030] Figure 6 A data communication flowchart for the training terminal provided in the embodiments of this application;

[0031] Figure 7 The human-factor intelligent prediction method provided for the implementation of this application;

[0032] Figure 8 A flowchart for determining a target prediction scheme from candidate prediction schemes, provided for an embodiment of this application;

[0033] Figure 9 A flowchart for obtaining human condition prediction results provided in an embodiment of this application;

[0034] Figure 10 A flowchart for obtaining human factor state prediction results based on human factor segment scores, provided for an embodiment of this application;

[0035] Figure 11 A schematic diagram of data communication at the prediction end provided for an embodiment of this application;

[0036] Figure 12 A schematic diagram of a human-computer interaction system provided for an embodiment of this application;

[0037] Figure 13 A schematic diagram of a human-computer interaction system provided for another embodiment of this application;

[0038] Figure 14 A schematic diagram of a training device for a human-centric intelligence model provided in an embodiment of this application;

[0039] Figure 15 A schematic diagram of a human factor intelligent prediction device provided for an embodiment of this application;

[0040] Figure 16 A block diagram of an edge computing device provided for an embodiment of this application. Detailed Implementation

[0041] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0042] Existing large-scale AI evaluation models are often trained on broad datasets, which makes it difficult to meet the application needs of specific vertical fields, and also fails to meet the functional requirements of human factors engineering in specific vertical fields, such as model training and state prediction.

[0043] Based on this, this application provides an open platform for human-centric intelligent assessment models, which empowers users with different backgrounds in AI algorithms across various fields. The open platform includes a training end and a prediction end. The training end provides automated operations throughout the entire process, from dataset creation, data source selection, data annotation, data denoising, model training, and model evaluation, allowing users to configure their own training schemes. The prediction end allows for flexible selection of single or multiple models for real-time prediction, and can also create comprehensive assessment systems with level-based alerts. Results support various visualization charts to help users intuitively understand and analyze data and assessment results, improving decision-making efficiency and accuracy.

[0044] The following provides a detailed description of the training end of the open platform for human factors intelligence models.

[0045] Figure 1 This is a flowchart of a human factors intelligence model training method according to an embodiment of this application.

[0046] As an example, such as Figure 1 As shown, the training methods for the human factors intelligence assessment model include S101-S103.

[0047] S101, Obtain candidate training schemes for the human factors intelligence model, wherein the candidate training schemes specify the training dataset and the model to be trained, wherein the training dataset is extracted from multiple human factors data sources according to training requirements.

[0048] S102, Based on the received selection instruction, determine the target training scheme from the candidate training schemes.

[0049] S103. Obtain the corresponding training dataset based on the target training scheme, and train the model to be trained based on the corresponding training dataset to obtain the trained model. The trained model can be used to predict human condition.

[0050] For example, the training end of the open platform for human factors intelligence models includes a list of training schemes, displaying multiple candidate training schemes. Each candidate training scheme specifies its training dataset and the model to be trained. Of course, if the desired scheme is not found in the list, the user can configure a candidate training scheme independently. The open platform includes massive human factors data sources, and the training dataset is extracted from these sources according to the user's training requirements. For instance, the open platform for human factors intelligence assessment models is connected to a human-machine environment synchronization platform, which can acquire massive amounts of human factors data. This platform includes various sensor devices, through which massive amounts of human factors data are collected.

[0051] For example, the training scheme list displays multiple candidate training schemes. Based on the user's selection instructions, the target training scheme is determined from the candidate schemes. The target training scheme also specifies the training dataset and the model to be trained. The training dataset is obtained from a massive human factors data source based on the training dataset specified in the target training scheme. Of course, before obtaining the training dataset, the platform also includes various preprocessing steps on the human factors data, such as data cleaning, data labeling, and data augmentation, to help users prepare and process the training data, improve the quality of the training data, and thus improve the training effect of the model, facilitating supervised learning model training. The model to be trained is then trained based on the obtained corresponding training dataset to obtain a trained model. The training process can call corresponding algorithms, and through hyperparameter optimization, find the optimal parameters for each algorithm corresponding to the training model to obtain a trained model. The trained model can be used to predict human factors states.

[0052] The training method of the human intelligence model in this application can be configured with various types of candidate training schemes according to the user's training needs, which is more flexible.

[0053] As an example, the training scheme list displays the names of multiple candidate training schemes. Each scheme in the list is unique; if two schemes have the same name, a distinguishing identifier will appear after the name. This training scheme list addresses the issues of repeatability and version management, ensuring the reproducibility of the training process and providing version management and traceability functions, making it convenient for users to manage and compare different versions of the model and experimental results.

[0054] As an example, such as Figure 2 As shown, before obtaining candidate training schemes for the human factors intelligence model, the training methods for the human factors intelligence model also include S201-S202.

[0055] S201, Obtain training scheme configuration information.

[0056] S202, Generate candidate training schemes based on training scheme configuration information.

[0057] For example, users can independently configure candidate training schemes through the open platform for human factors intelligent models of this application. The open platform can output a training scheme configuration interface on the front end. Users can select or input information on the training scheme configuration interface, and the background obtains the training scheme configuration information and generates candidate training schemes based on the training scheme configuration information.

[0058] For example, the training scheme configuration interface may include multiple configuration controls, which are used to provide users with input configuration information. Candidate training schemes are generated based on the training scheme configuration information received from the training scheme configuration interface. The candidate training schemes specify the training dataset and the model to be trained.

[0059] As an example, such as Figure 3 As shown, candidate training schemes, including S301-S302, are generated based on the training scheme configuration information.

[0060] S301, obtain training data source configuration information.

[0061] S302: Based on the training data source configuration information, obtain the corresponding data from the human-machine environment synchronization platform, process the obtained data, and use the processed data as the training dataset for candidate training schemes.

[0062] For example, the training scheme configuration information includes training data source configuration information, and the training scheme configuration interface may also include a training data source configuration control. Users can input or select information in the training data source configuration control, and the background obtains the training data source configuration information. Based on the training data source configuration information, the corresponding data is obtained from the human-machine environment synchronization platform.

[0063] For example, such as Figure 4 The diagram shown illustrates the training scheme configuration interface. The interface includes controls for configuring training data sources, for example... Figure 4 The data source selection section. It retrieves training data source configuration information based on the training data source configuration control, for example, such as... Figure 4 As shown, users can select either Heart Rate Variability (HRV) or Electroencephalogram (EEG) data sources, or both. Different interfaces may offer other data source options, such as infrared data. Based on the training data source configuration information, corresponding data is retrieved from a massive backend dataset. The Human Factors Intelligent Model Open Platform communicates with the Human-Machine Environment Synchronization Platform, which supports simultaneous acquisition from different types of human physiological recorders. The Human-Machine Environment Synchronization Platform collects a large amount of data related to the training data source configuration information and performs preprocessing operations such as cleaning, deduplication, noise reduction, and missing value imputation to ensure data quality. The Human-Machine Environment Synchronization Platform has no limit on the number of physiological sensors and signals simultaneously acquired in the same recording, supporting data acquisition and transmission for cloud projects. The data is saved as files on the platform for project use. The processed data is used as the training dataset for candidate training schemes.

[0064] As an example, training data sources include at least one of the heart rate data source HRV and the electroencephalogram (EEG) data source.

[0065] For example, such as Figure 4As shown, the data source can be selected from heart rate data source HRV, brain wave data source EEG, or both. Under heart rate data source HRV, you can further select at least one of ECG (electrocardiogram) and PPG (pulse photogrammetry) signals. Under brain wave data source EEG, you can further select the signal channel data, for example, 8 channels or 16 channels.

[0066] As an example, training data sources include at least one of the following: human-related data sources, machine-related data sources, human-computer interaction-related data sources, and environment-related data sources;

[0067] Among them, human-related data sources include one or a combination of the following: skin conductance and temperature data, pulse data, blood pressure data, blood oxygen data, electrocardiogram data, electromyography data, muscle oxygen data, respiratory data, biomechanical data, near-infrared brain imaging data, electroencephalogram data, transcranial stimulation data, heart rate variability data, heart rate data, image / video data, sound data, eye movement data, gesture or movement data.

[0068] Machine-related data sources include one or a combination of the following: machine operation data, fault alarm data, machine control data, machine model data, machine communication data, and machine positioning data;

[0069] Human-computer interaction related data sources include one or a combination of the following: human-computer voice interaction data, human-computer text interaction data, human-computer touch interaction data, human-computer gesture or motion interaction data, human-computer EEG interaction data, human-computer eye-tracking interaction data, and human-computer facial expression interaction data;

[0070] Environmental data sources include one or a combination of the following: location data, humidity data, temperature data, colorimetric data, brightness data, weather data, road condition data, traffic data, stimulus signal data, and event or signal tagging data.

[0071] As an example, the training data source configuration information includes baseline data configuration information and non-baseline data configuration information. Based on the training data source configuration information, the corresponding data is obtained from the human-machine environment synchronization platform, including: obtaining baseline data from the human-machine environment synchronization platform based on the baseline data configuration information, and obtaining non-baseline data from the human-machine environment synchronization platform based on the non-baseline data configuration information. The training dataset includes baseline data and non-baseline data, and / or, the baseline data is used to verify or optimize the performance of the human factors intelligence model.

[0072] For example, the training data source configuration information includes baseline data configuration information and non-baseline data configuration information. Users can enter or select whether to configure baseline data in the training scheme configuration interface. When the baseline data configuration information indicates that baseline data needs to be configured, the background obtains the baseline data from the human-machine environment synchronization platform according to the baseline data configuration information. It can also obtain non-baseline data from the human-machine environment synchronization platform according to the non-baseline data configuration information. The training dataset includes baseline data and non-baseline data. The baseline data is used to verify or optimize the performance of the human factors intelligence model.

[0073] For example, such as Figure 4 As shown, in the data source selection section, users can also choose whether to use a baseline. Baseline data refers to physiological data signals collected over a period of time before the experiment begins, under a resting state, and is used as a reference for subsequent experimental data. Baseline data helps researchers understand an individual's normal physiological state, thereby better analyzing changes during the experimental process.

[0074] As an example, obtaining the corresponding data from the human-machine environment synchronization platform includes: reading the configuration file created in the system background to obtain the corresponding data from the human-machine environment synchronization platform, wherein the configuration file includes information used to characterize the data source.

[0075] For example, after the user selects the data source in the training data source configuration control on the training scheme configuration interface, the system background creates a configuration file. The human-machine environment synchronization platform reads the configuration file to obtain the data corresponding to the training data source information. The configuration file includes information used to characterize the data source. Initially, the system background has a certain data source. The system background can create a configuration file and read the configuration file to obtain the data source. This process can be understood as the process by which the system obtains human factors data collected by the acquisition device by reading the configuration file in the background.

[0076] For example, after configuring the training scheme in the configuration interface, click "Save and Train". The background will launch the data acquisition device according to the data source selection in the training scheme configuration interface. However, there is another option here, which allows you to select a data source for a specific item, such as selecting data under the eye tracker device item.

[0077] As an example, the information used to characterize the data source includes information about the acquisition device, which includes at least one of the following: smart wearable sensors, biosensors, mobile portable sensors, and eye trackers.

[0078] For example, the human-machine environment synchronization platform is connected to the human factors intelligent assessment platform, supporting the synchronous acquisition of data from different types of human physiological recorders, including at least one of smart wearable sensors, biosensors, mobile portable sensors, and eye trackers. The human-machine environment synchronization platform reads the configuration file to obtain the data from the acquisition devices, and after preprocessing operations such as cleaning, deduplication, noise reduction, and filling in missing values, it saves the data in file form.

[0079] For example, the exported data will be stored in a local folder, and the folder path can be determined by... Figure 4 The directory path is determined. Data consistency is crucial for ensuring the repeatability and reliability of model training. By saving the training dataset, it's ensured that all team members use the same file and the same data version, avoiding errors and biases caused by data inconsistencies. Saving the training dataset allows the data to be reused by multiple projects or teams, thereby improving work efficiency and resource utilization. Alternatively, establishing a dataset repository can also promote data reuse and sharing.

[0080] As an example, such as Figure 5 As shown, candidate training schemes are generated based on training scheme configuration information, and S501-S502 are also included.

[0081] S501, Obtain information about the type of the model to be trained.

[0082] S502, Based on the type information of the model to be trained, determine the algorithm model corresponding to the type information of the model to be trained from the candidate algorithm models, and use it as the model to be trained for the candidate training scheme.

[0083] For example, such as Figure 4 The algorithm settings and training scheme configuration interface shown also includes a control for configuring the type of model to be trained. Users can select the desired model algorithm from this control, choosing from multiple or all options, or selecting combinations of algorithms. The backend retrieves the type information of the model to be trained based on this configuration control. Then, it determines the corresponding algorithm model from the candidate algorithm models based on this type information, using it as the training model for the candidate training scheme.

[0084] As an example, candidate algorithm models include at least one of the following: linear discriminant analysis model, random forest model, support vector machine algorithm model, logistic regression model, decision tree model, and classification algorithm model.

[0085] For example, the open platform for human-computer intelligent models in this application supports various machine learning algorithms, such as Linear Discriminant Analysis (LDA), Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree (DT), and KNN classification algorithms. Users can choose at least one of these for model training. Specifically, for LDA, RF, and SVM, the platform also provides LDA based on Filter Bank Common Space Pattern (FBCSP+LDA), RF based on Filter Bank Common Space Pattern (FBCSP+RF), and SVM based on Filter Bank Common Space Pattern (FBCSP+SVM). It provides abundant tools and resources to support the development and optimization of machine learning models. It employs deep machine learning AI algorithms, such as using features from EEG and PPG signals, by inputting data into the model and iteratively adjusting the model's parameters to gradually approach the optimal solution, enabling it to accurately perform prediction or classification tasks, or enabling the model to learn features and patterns from data and improve its performance.

[0086] As an example, the acquired data includes human factors segments. The data acquired from the backend is processed, including: acquiring tag information, where the tag information is used to label the state of the human factors segments; and adding the tag information to the human factors segments.

[0087] For example, the training scheme configuration interface also includes a training data label configuration control. Users can add the label information required for the experiment in the training data label configuration control, and the backend can obtain the label information based on the training data label configuration control. Label information can be manually added to the human factor segments. The label information is used to label the state of the human factor segments. It can be understood that the label information is used to label the training dataset. The model to be trained is trained based on the training dataset. The model to be trained performs operations such as feature extraction on the training dataset to obtain the predicted labels of the training dataset. The predicted labels of the training dataset can be compared with the pre-added label information, and the model to be trained can be continuously trained and iterated to make the labels predicted by the model to be trained closer to the pre-added label information.

[0088] For example, such as Figure 4 The label configuration controls A, B, C, and D shown are provided. The number of label configuration controls is not limited to four; other numbers are also allowed, but at least two labels are required. Labels A, B, C, and D are manually filled in by the user according to the experiment type. For example, if the user wants to assess stress emotions, they can add labels such as high load, medium load, and low load. The training protocol configuration interface also includes a protocol name control, which can be filled in by the user. Figure 4After filling in the information on the training scheme configuration interface, click "Save and Train". The human factors intelligent assessment model platform exports data based on the training data source information, and this data can be further labeled. The data obtained in the background includes human factors segments. Select any one of the A, B, C, or D labels entered in the label configuration control to label the human factors segment. This is understandable. Figure 4 The tags A, B, C, and D are the tags the user intends to assign. Figure 4 After configuring, click "Save and Train," and the backend will then... Figure 4 Selecting a data source will bring up the data acquisition device list. For example, selecting the eye tracker as the device will acquire the eye tracker data. After that, you can manually label the fragments that match the specified type from the eye tracker data. It's understandable that fragments matching the specified type are those that can be labeled A, B, C, and D. The eye tracker data will certainly contain some FGH fragments, which are not the fragments needed for the experiment. Figure 4 Select one label from all labels A, B, C, and D, and then click the export button to obtain the training dataset for training purposes.

[0089] It's worth noting that if you select to use a baseline when configuring the data source, an additional baseline option will appear when labeling. For example, when the option to use a baseline is selected, you can assign a baseline label to a specific segment for use by subsequent training and prediction algorithms.

[0090] The model to be trained is trained using the training dataset. The parameters of the model are iteratively adjusted repeatedly to gradually approach the optimal solution, resulting in a well-trained model.

[0091] As an example, training scheme management includes a list of training schemes, displaying candidate training schemes. If the desired training method is already among the candidates, it can be selected directly. Otherwise, the user can create a new training scheme, configuring the scheme name, selecting the training data source and baseline, setting labels, and selecting the algorithm model to complete the configuration. The training scheme is saved in a folder within the scheme storage path, and training begins. The acquired data is labeled according to the label settings for training purposes. The training end includes data collection, data preprocessing, model training, and optimization. In practical applications, targeted technical challenges can be implemented based on specific needs and data characteristics to improve system performance and effectiveness. Training scheme management allows users to open or create schemes and, through the training scheme parameter configuration process, provides configuration information (config) such as the data source, labels, and algorithms used for model training. Data and label information are provided through project data management. The training-side configuration solution aims to address issues such as data preparation, model design, and hyperparameter tuning during deep learning model training. Based on the selected data source, labels, and data, it performs specific data denoising and feature extraction, creates training sets, calls corresponding algorithms, and finds the optimal parameters for each algorithm's corresponding model through hyperparameter optimization. The training end supports training on large amounts of data, revealing patterns, trends, and correlations within a set of models, thereby providing insights and support for decision-making and business process optimization. This allows for the selection of the optimal model architecture. Thus, it provides rich, flexible, and efficient targeted solutions.

[0092] During model training, a training progress bar can be displayed, and training can be terminated during the process to reconfigure data sources, labels, algorithms, and other configuration information.

[0093] As an example, the preprocessing process includes filtering, signal quality detection (for verifying frequency domain proportion), amplitude rule, IBI extraction, abnormal IBI point detection and correction, and feature extraction based on the preprocessing process. Extractable features include time-domain MeanIBI, meanHR, frequency-domain ULF, VLF, and nonlinearity. The extracted features are further standardized, dimensionality reduced, and filtered to obtain the optimal feature subset. The algorithm selected in the scheme configuration is used to train a model using a machine learning framework for prediction.

[0094] Figure 6 This is a data communication flowchart of the training end according to an embodiment of this application.

[0095] like Figure 6As shown, the human factors intelligence platform creates a training plan and configures data sources and algorithms. The data source collection system exports the cleaned and labeled dataset (i.e., the training dataset) based on the data annotations. After the data is ready, the human factors intelligence platform provides the signal sources currently available on the platform (e.g., PPG, ECG) and has developed corresponding algorithms and recommended algorithms for each signal source (each signal source has its own recommended algorithm). Users can make personalized choices on the data sources and algorithms provided by the system based on the experimental scenario (e.g., evaluating EEG or oculomotor signals) and design, in order to select a suitable model architecture and algorithm for training. The human factors intelligence evaluation system starts the machine learning algorithm to train according to the training configuration parameters and returns the training progress in real time via communication protocol to display it to the human factors intelligence evaluation system. After training, the model needs to be evaluated to measure its performance. Validation sets or cross-validation are typically used to evaluate the model's performance on unseen data. Based on the evaluation results, model tuning may be necessary, including adjusting hyperparameters and improving feature engineering, to preserve the optimal model.

[0096] This application also proposes a human-centric intelligent prediction method.

[0097] As an example, such as Figure 7 As shown, the human factors intelligence assessment methods include S701-S703.

[0098] S701 extracts the dataset to be predicted from multiple human factors data sources.

[0099] S702, based on the received prediction scheme selection operation, determines the target prediction scheme from the candidate prediction schemes, wherein the target prediction scheme specifies the trained model.

[0100] S703, based on the target prediction scheme, uses a trained model to predict the human condition of the dataset to be predicted.

[0101] For example, after training the model using the aforementioned human factors intelligence model training method, the prediction end of the human factors intelligence platform predicts human factors data. The human factors intelligence assessment platform is connected to a human-machine environment synchronization platform, enabling it to collect human factors data in real time via acquisition devices and extract the dataset to be predicted from the human factors data source. The prediction end also includes a prediction scheme list, displaying multiple candidate prediction schemes. Based on the user's selection, a target prediction scheme is determined from the candidate schemes. The target prediction scheme includes the model trained using the aforementioned training method. Based on the target prediction scheme, the trained model is used to predict the human factors state of the dataset to be predicted, obtaining the human factors state prediction result.

[0102] For example, the target prediction scheme specifies a trained model, which can be a model trained according to the above-mentioned human factors intelligence model training method, or a model that the user has trained elsewhere and then imported into the human factors intelligence platform for prediction or retraining and optimization.

[0103] For example, the created prediction schemes are displayed in a prediction scheme list. The model is updated regularly to adapt to new data and scenarios, maintaining stable and optimized prediction performance. Furthermore, the prediction scheme list facilitates user management and comparison of different versions, allowing for easy review and comparison of their effectiveness.

[0104] As an example, such as Figure 8 As shown, based on the received prediction scheme selection operation, the target prediction scheme is determined from the candidate prediction schemes, including S801-S802.

[0105] S801, Obtain prediction scheme configuration information, which includes target prediction scheme path information.

[0106] S802, determine the target prediction scheme from the candidate prediction schemes based on the target prediction scheme path information.

[0107] For example, the prediction endpoint outputs a prediction scheme configuration interface, where users can configure prediction schemes. The backend retrieves the prediction scheme configuration information from the interface, including the target prediction scheme path information. The target prediction scheme is then determined from the candidate prediction schemes based on this path information. For instance, the target prediction scheme can be directly retrieved via its path.

[0108] For example, the prediction endpoint can output a prediction scheme configuration interface. Users can enter the prediction scheme name in the configuration interface and also select to add a prediction scheme from the training scheme list. The prediction scheme configuration interface includes a target prediction scheme path control, through which the target prediction scheme path information can be obtained. The target prediction scheme path is the storage path of the trained model. Based on the selected target prediction scheme path information, the target prediction scheme is determined from the candidate prediction schemes. It can be understood that the trained model is stored in a file; there are multiple models. During prediction, a model is selected from the file and imported into the prediction endpoint. The algorithm provided by the model is then selected for prediction. The training endpoint can select multiple algorithms for training, while the prediction endpoint can select one algorithm for prediction.

[0109] Of course, the default model and default prediction scheme provided by the human factors intelligence platform in this application are derived from customer-customized models and some classic models that we have developed in our own way.

[0110] As an example, such as Figure 9 As shown, the trained model is used to predict the human condition of the dataset to be predicted, including S901-S902.

[0111] S901 uses the trained model to predict the dataset to be predicted, and obtains the label of each person's factor segment.

[0112] S902, score the human factor segment based on the label score corresponding to the predicted label to obtain the score of the human factor segment, and obtain the human factor state prediction result based on the score of the human factor segment.

[0113] For example, the dataset to be predicted includes at least one human factor segment, and the prediction scheme configuration interface also includes a score configuration control, which is used to receive configured label scores. For example, the user can customize the label scores in the score configuration control interface (e.g., 10 points for high load, 20 points for medium load, and 30 points for low load). The trained model is used to predict the dataset to be predicted, and the label for each human factor segment is predicted. The human factor segment is scored according to the label score corresponding to the label to obtain a score for the human factor segment. The human factor state prediction result is obtained based on the score of the human factor segment. For example, a higher score indicates a more relaxed human factor state.

[0114] As an example, such as Figure 10 As shown, each human factor segment is scored based on the corresponding score value of the label, and the human factor state prediction result is obtained based on the human factor segment score, including S1001-S1002.

[0115] S1001, determine the comprehensive score based on the scores of multiple human factor segments, wherein the comprehensive score is obtained by weighting the tag scores corresponding to the multiple human factor segments.

[0116] S1002, based on the warning level configuration information and comprehensive score value, the human factor status prediction result is obtained.

[0117] For example, the prediction dataset includes multiple human factor segments, and the prediction scheme configuration interface also includes a warning level configuration control, which receives warning level configuration information. As an indicator, the label score can be combined with existing indicators to generate new indicators for more diversified evaluation. Users can configure warning levels and warning conditions in the comprehensive evaluation indicator configuration interface. For instance, if a user has three algorithm models—fatigue, load, and stress—and wants to continuously evaluate a pilot's state information while completing a task, they can assign different weights to fatigue, load, and stress, and combine them using a formula to generate a new comprehensive performance indicator (i.e., a comprehensive score). The warning level configuration control receives warning level configuration information, which may include each warning level and the corresponding warning conditions. The human factor state prediction result is obtained based on the warning level configuration information and the comprehensive score. For example, an alarm may be triggered when the high load score is highest, and the alarm also has corresponding levels.

[0118] As an example, users can train and optimize models through the human factors intelligence system, i.e., import the training scheme they want to use (i.e., the process of importing the trained model into the prediction end as described above). This is the preparation process in the prediction process. As needed, the algorithm can be selected, and the platform can customize the label score. Users can calculate the score required for the experiment (which can be a comprehensive score with weighted label scores) based on the experimental design and experimental scenario through expert evaluation methods, and fill it into the platform's customized label score to obtain the best prediction performance.

[0119] As an example, the human factors status assessment results include a warning level. Based on the warning level configuration information and the comprehensive score, the human factors status prediction results are obtained, including: if the comprehensive score meets the warning conditions, the warning level is determined based on the warning level configuration information.

[0120] For example, the prediction scheme configuration interface also includes the creation of an evaluation and early warning system. The comprehensive indicators, their names, calculation formulas, and early warning levels can be configured in the comprehensive indicator interface (for example, in the comprehensive indicators, a high load of 50 points represents early warning level 1, and a high load of 80 points represents early warning level 2), and then prediction is performed.

[0121] As an example, the prediction scheme list allows users to choose a default prediction scheme or add a new one. If adding a prediction scheme, clicking the "Add" button allows users to select a path (i.e., the path to the trained model), customize labels and assign label values ​​(which can be default scores or custom scores), and then click the "Create" button to create a comprehensive index, its name, calculation formula, and warning level. After completing the configuration, prediction can begin. Clicking "Save and Predict" after configuring the prediction scheme will activate the data source acquisition device (e.g., an eye tracker). If data is unavailable, the user may be prompted to connect the device, and the status will be refreshed after connection to obtain the prediction dataset. Users select the necessary device signal sources (to obtain the prediction dataset, i.e., which data to predict) and, combined with the configuration of the human factors intelligence system's prediction function, use the machine learning algorithm module to predict the results. The human factors intelligence platform may utilize various machine learning algorithms, including supervised learning, unsupervised learning, and reinforcement learning, to train prediction models. These models can be used to predict future trends, behaviors, or events. Optimization of the model and inference process improves prediction speed and efficiency. Techniques such as model compression, quantization, and pruning can be used to reduce model size and computational cost. Post-processing of the model output may include result decoding, post-processing filtering, and other operations to improve the accuracy and stability of the prediction results.

[0122] Figure 11 This is a schematic diagram of data communication at the prediction end according to an embodiment of this application.

[0123] like Figure 11 As shown, Figure 11The process of converting state data into encoding can be understood as the human factors intelligence platform converting the data it needs into an encoding, which the human-machine environment synchronization platform then uses to collect data. The human factors intelligence assessment system receives raw signals from the data source acquisition system and preprocesses them. For HRV signals, to eliminate noise interference from environmental factors and motion, this application performs several processing steps. For example, it first applies a 1Hz high-pass filter to the raw signal, then verifies its frequency domain proportion and amplitude regularity, calculates the envelope to obtain the R-Peak, verifies the start and end times of the R-Peak, calculates the variance, verifies stability, performs IBI interpolation and IBI spectrum, verifies whether the IBI time difference matches a normal heartbeat, and also performs time-domain, frequency-domain, and nonlinear feature extraction. For EEG signals, a self-developed FFT filter is used, followed by Fast Fourier Transform to calculate the PSD value, extracting frequency domain features, etc., to obtain the optimal feature subset. This process can be understood as a series of processes, including removing interference from the prediction dataset, to obtain the processed prediction dataset. The prediction dataset is then transmitted in real-time to the pre-trained model for prediction and inference. AI prediction can be performed through two different algorithm modules. For example, Machine Algorithm 1 performs complex calculations or algorithmic processing and returns the prediction result. Machine Algorithm 2 performs machine learning-related processing tasks and returns the prediction result. The prediction result from the machine learning algorithm module is returned to the human factors intelligent assessment platform. The human factors intelligent assessment platform receives the prediction result and performs matching processing with the set labels. It can be understood that the human factors data to be predicted was originally labeled; the predicted dataset labels are compared with the true labels. The processed assessment results are then transmitted back to the human factors intelligent assessment platform. For example, the assessment results include information such as the comparison between the true labels and the predictions, and the model's prediction success rate. The human factors intelligent assessment platform can also send the processed data to the UI for visualization via a communication protocol.

[0124] This application also proposes a human-centric intelligent system.

[0125] As an example, such as Figure 12 As shown, the human factors intelligence system 100 includes a human factors intelligence platform 10, which includes a training end and a prediction end; the training end is used to execute the human factors intelligence model training method described above; the prediction end is used to execute the human factors intelligence prediction method described above.

[0126] For example, the human factors intelligence platform 10 includes a training end and a prediction end. The training end can provide automated operations throughout the entire process, from dataset creation, data source selection, data annotation, data denoising, model training, and model evaluation. Users can configure the training scheme independently. The prediction end can flexibly select single or multiple models for real-time prediction, and can also create a comprehensive evaluation system and provide level-based early warnings. The results support various visualization charts to help users intuitively understand and analyze data and evaluation results, thereby improving decision-making efficiency and accuracy.

[0127] The human factors intelligence platform of this application has accumulated a large amount of data and rich industry experience in the field of human factors engineering. It can provide source collection of various data, empower users with different AI algorithm foundations in different fields, provide customized AI models, and realize the function of multimodal data synchronous collection and analysis module.

[0128] As an example, such as Figure 13 As shown, the human factors intelligent system 100 also includes a human-machine environment synchronization platform 20. The human factors intelligent platform 10 is connected to the human-machine environment synchronization platform 20. The human-machine environment synchronization platform 20 is used to store, acquire and / or process human factors data collected by human factors signal acquisition devices.

[0129] For example, the human-machine environment synchronization platform 20 can acquire human factors data through various types of acquisition devices and perform various processing on the human factors data, such as data cleaning, data annotation, and data augmentation, to help users prepare and process data and improve data quality. The human-machine environment synchronization platform 20 is connected to the human factors intelligent platform 10 and receives instructions from the human factors intelligent platform 10 to provide the human factors intelligent platform 10 with the human factors data required.

[0130] The training and prediction data in this application were both collected through a human-machine environment synchronization platform. This platform acquires a wealth of human-factor environment data through human-factor signal acquisition equipment. Reusing this data, combined with a human-factor intelligence platform, can provide customizable, rich, and flexible artificial intelligence model training and prediction solutions for the human-factor intelligence field.

[0131] This application proposes a training device for human-caused intelligence models.

[0132] As an example, such as Figure 14 As shown, the training device for the human-computer interaction model includes:

[0133] The acquisition module 1401 is used to acquire candidate training schemes for the human factors intelligence model. The candidate training schemes specify the training dataset and the model to be trained. The training dataset is extracted from multiple human factors data sources according to training requirements. The determination module 1402 is used to determine the target training scheme from the candidate training schemes based on the received selection instructions. The training module 1403 is used to acquire the corresponding training dataset based on the target training scheme and train the model to be trained based on the corresponding training dataset to obtain a trained model. The trained model can be used to predict human factors state.

[0134] This application proposes a human-caused intelligent prediction device.

[0135] As an example, such as Figure 15 As shown, the human factors intelligent prediction device includes: an extraction module 1501, used to extract the dataset to be predicted from multiple human factors data sources; a determination module 1502, used to determine the target prediction scheme from the candidate prediction schemes based on the received prediction scheme selection operation, wherein the target prediction scheme specifies a trained model; and a prediction module 1503, used to predict the human factors state of the dataset to be predicted based on the target prediction scheme and using the trained model.

[0136] This application also proposes a computer-readable storage medium.

[0137] In this embodiment, a computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the above-described human factors intelligence model training and prediction method.

[0138] Figure 16 A block diagram of an edge computing device provided for an embodiment of this application.

[0139] This application provides an edge computing device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described human factors intelligence model training and prediction method.

[0140] like Figure 16 As shown, for ease of understanding, embodiments of this application illustrate a specific edge computing device.

[0141] Edge computing devices are intended to represent various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Edge computing devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0142] like Figure 16 As shown, the device includes a computing unit 1601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 1602 or a computer program loaded into random access memory (RAM) 1603 from storage unit 1608. The RAM 1603 may also store various programs and data required for the operation of the edge computing device. The computing unit 1601, ROM 1602, and RAM 1603 are interconnected via bus 1604. An input / output (I / O) interface 1605 is also connected to bus 1604.

[0143] Multiple components in the edge computing device are connected to I / O interface 1605. These components include: input units 1606, such as a keyboard and mouse; output units 1607, such as various types of displays and speakers; storage units 1608, such as disks and optical discs; and communication units 1609, such as network interface cards (NICs), modems, and wireless transceivers. Communication unit 1609 allows the edge computing device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0144] The computing unit 1601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1601 performs the various methods described above, such as human-cause intelligence model training and prediction methods. For example, in some embodiments, the human-cause intelligence model training and prediction methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 1608. In some embodiments, part or all of the computer program may be loaded into and / or installed on an edge computing device via ROM 1602 and / or communication unit 1609. When the computer program is loaded into RAM 1603 and executed by the computing unit 1601, the human-cause intelligence model training and prediction methods described above can be performed. Alternatively, in other embodiments, computing unit 1601 may be configured to perform human factor intelligence model training and prediction methods by any other suitable means (e.g., by means of firmware).

[0145] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this application, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0146] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0147] In the description of this application, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0148] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicating the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0149] Furthermore, the terms "first," "second," etc., used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance, or implicitly specifying the number of technical features indicated in this embodiment. Therefore, features defined with terms such as "first" and "second" in the embodiments of this application can explicitly or implicitly indicate that the embodiment includes at least one of those features. In the description of this application, the word "multiple" means at least two or more, such as two, three, four, etc., unless otherwise explicitly and specifically defined in the embodiments.

[0150] In this application, unless otherwise explicitly specified or limited in the embodiments, the terms "installation," "connection," "joining," and "fixing" appearing in the embodiments should be interpreted broadly. For example, a connection can be a fixed connection, a detachable connection, or an integral part; it can also be a mechanical connection, an electrical connection, etc. Of course, it can also be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication between two components, or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific implementation.

[0151] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0152] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for training a human-centric intelligent model, characterized in that, The method includes: Obtain candidate training schemes for human factors intelligence models, wherein the candidate training schemes specify a training dataset and a model to be trained, wherein the training dataset is extracted from multiple human factors data sources according to training requirements; Based on the received selection instruction, a target training scheme is determined from the candidate training schemes; The corresponding training dataset is obtained based on the target training scheme, and the model to be trained is trained based on the corresponding training dataset to obtain a trained model. The trained model can be used to predict human condition.

2. The human-centric intelligent model training method according to claim 1, characterized in that, Before obtaining candidate training schemes for the human-centric intelligence model, the method further includes: Obtain training scheme configuration information; Candidate training schemes are generated based on the training scheme configuration information.

3. The human-centric intelligent model training method according to claim 2, characterized in that, The training scheme configuration information includes training data source configuration information, and the generation of candidate training schemes based on the training scheme configuration information includes: Obtain training data source configuration information; Based on the training data source configuration information, corresponding data is obtained from the human-machine environment synchronization platform, and the obtained data is processed. The processed data is used as the training dataset for the candidate training scheme.

4. The human-centric intelligent model training method according to claim 2, characterized in that, The step of generating candidate training schemes based on the training scheme configuration information includes: Obtain information about the type of the model to be trained; Based on the type information of the model to be trained, an algorithm model corresponding to the type information of the model to be trained is determined from the candidate algorithm models, and used as the model to be trained for the candidate training scheme.

5. The human-centric intelligent model training method according to claim 3, characterized in that, The acquired data includes human factor segments, and the processing of the acquired data includes: Obtain tag information, wherein the tag information is used to label the state of the human factor segment; Add the tag information to the human factors segment.

6. The human-centric intelligent model training method according to claim 3, characterized in that, The process of obtaining the corresponding data from the human-machine environment synchronization platform includes: The system reads a configuration file created in the background to obtain corresponding data from the human-machine environment synchronization platform. The configuration file includes information used to characterize the source of the data.

7. The human-centric intelligent model training method according to claim 3, characterized in that, The training data source configuration information includes baseline data configuration information and non-baseline data configuration information. The step of obtaining corresponding data from the human-machine environment synchronization platform based on the training data source configuration information includes: Baseline data is obtained from the human-machine environment synchronization platform based on the baseline data configuration information, and non-baseline data is obtained from the human-machine environment synchronization platform based on the non-baseline data configuration information, wherein the training dataset includes the baseline data and the non-baseline data, and / or the baseline data is used to verify or optimize the performance of the human factors intelligence model.

8. The human-centric intelligent model training method according to claim 6, characterized in that, Information used to characterize the data source includes information about the acquisition device, which includes at least one of the following: a smart wearable sensor, a biosensor, a mobile portable sensor, an eye tracker; and / or, The training data sources include at least one of the following: human-related data sources, machine-related data sources, human-computer interaction-related data sources, and environment-related data sources; wherein, the human-related data sources include one or a combination of the following: skin conductance and temperature data, pulse data, blood pressure data, blood oxygen data, electrocardiogram data, electromyography data, muscle oxygenation data, respiratory data, biomechanical data, near-infrared brain imaging data, electroencephalogram data, transcranial stimulation data, heart rate variability data, heart rate data, image / video data, sound data, eye-tracking data, and gesture or movement data; the machine-related data sources include one or a combination of the following: machine movement... The data sources include: line data, fault alarm data, machine control data, machine model data, machine communication data, and machine positioning data; the human-computer interaction related data sources include one or a combination of the following: human-computer voice interaction data, human-computer text interaction data, human-computer touch interaction data, human-computer gesture or action interaction data, human-computer EEG interaction data, human-computer eye-tracking interaction data, and human-computer facial expression interaction data; the environment related data sources include one or a combination of the following: location data, humidity data, temperature data, color data, brightness data, weather data, road condition data, traffic data, stimulus signal data, and event or signal tagging data; and / or, Candidate algorithm models include at least one of the following: linear discriminant analysis model, random forest model, support vector machine algorithm model, logistic regression model, decision tree model, and classification algorithm model.

9. A human-centric intelligent prediction method, characterized in that, The method includes: Extract the dataset to be predicted from multiple human factors data sources; Based on the received prediction scheme selection operation, a target prediction scheme is determined from the candidate prediction schemes, wherein the target prediction scheme specifies a trained model; Based on the target prediction scheme, the trained model is used to predict the human condition of the dataset to be predicted.

10. The human-cause intelligent prediction method according to claim 9, characterized in that, The step of selecting a target prediction scheme from candidate prediction schemes based on the received prediction scheme selection operation includes: Obtain prediction scheme configuration information, wherein the prediction scheme configuration information includes target prediction scheme path information; The target prediction scheme is determined from the candidate prediction schemes based on the path information of the target prediction scheme.

11. The human-cause intelligent prediction method according to claim 10, characterized in that, The dataset to be predicted includes at least one human factor segment, and the step of using the trained model to predict the human factor state of the dataset to be predicted includes: The trained model is used to predict the dataset to be predicted, and the label of each individual fragment is predicted. The human factor segment is scored based on the label score corresponding to the predicted label, and the human factor state prediction result is obtained based on the score of the human factor segment.

12. The human-cause intelligent prediction method according to claim 11, characterized in that, The human factor segments include multiple segments. The step of scoring each human factor segment based on the tag score corresponding to the tag, and obtaining a human factor state prediction result based on the human factor segment score, includes: A comprehensive score is determined based on the scores of the multiple human factor segments, wherein the comprehensive score is obtained by weighting the tag scores corresponding to the multiple human factor segments; The human condition prediction result is obtained based on the warning level configuration information and the comprehensive score.

13. The human-cause intelligent prediction method according to claim 12, characterized in that, The human condition prediction result includes a warning level. The process of obtaining the human condition prediction result based on the warning level configuration information and the comprehensive score includes: If the comprehensive score meets the warning conditions, the warning level is determined according to the warning level configuration information.

14. A human-centric intelligent system, characterized in that, The human-centric intelligent system includes a human-centric intelligent platform, which includes a training end and a prediction end. The training terminal is used to execute the human factors intelligent model training method as described in any one of claims 1-8; The prediction segment is used to perform the human-caused intelligent prediction method as described in any one of claims 9-13.

15. The human-centric intelligent system according to claim 14, characterized in that, The human factors intelligent system also includes a human-machine environment synchronization platform, which is connected to the human-machine environment synchronization platform. The human-machine environment synchronization platform is used to store, acquire, and / or process human factors data collected by human factors signal acquisition devices.

16. An edge computing device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the training steps of the human factors intelligence model corresponding to the method of any one of claims 1-8 and / or the human factors intelligence prediction steps corresponding to the method of any one of claims 9-13.

17. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the training steps of the human-cause intelligence model corresponding to the method of any one of claims 1-8 and / or the human-cause intelligence prediction steps corresponding to the method of any one of claims 9-13.

Images