Multi-level and multi-field model assembly and fusion method and system

Abstract

The invention discloses a multi-level and multi-field model assembly and fusion method and system, which adopt an advanced data fusion algorithm to fuse data from different sensors, improve the accuracy and comprehensiveness of the data and provide a more reliable basis for the construction of a digital twinborn model. Machine learning and deep learning algorithms are utilized to construct a data-driven digital twinborn model, so that the future state and performance of the equipment can be accurately predicted, and powerful support is provided for intelligent management and maintenance of the urban rail field equipment; and a set of complete virtual-real consistency verification index system and verification method is designed, so that the accuracy and reliability of the digital twin model can be effectively ensured, and a guarantee is provided for the application of the digital twin technology in the urban rail field.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a method and system for assembling and fusing multi-level and multi-domain models. Background Art

[0002] With the rapid development of artificial intelligence technology, various models have been widely applied in different fields. However, a single model often fails to meet the requirements of complex problems and has problems such as limited performance and insufficient generalization ability. In order to improve the performance and generalization ability of models, researchers have begun to explore methods for model assembly and fusion. Currently, some model assembly and fusion methods have been proposed, but these methods have some deficiencies. For example, some methods are only applicable to specific fields or specific types of models and lack generality; the assembly and fusion processes of some methods are relatively complex and have high computational costs; the performance improvement of some methods is limited and it is difficult to meet the requirements of practical applications. Therefore, a new multi-level and multi-domain model assembly and fusion method is needed, which can overcome the deficiencies of existing methods and improve the performance and generalization ability of models. Summary of the Invention

[0003] This application provides a method and system for assembling and fusing multi-level and multi-domain models to solve the problems of the existing model assembly and fusion methods, such as lack of generality, high cost, limited performance improvement, and difficulty in meeting the requirements of practical applications.

[0004] According to a first aspect, in one embodiment, a method for assembling and fusing multi-level and multi-domain models is provided. The method includes:

[0005] Select a suitable model according to specific application requirements and problem characteristics;

[0006] Divide the selected models into different levels;

[0007] Perform domain adaptation for models in different fields;

[0008] Assemble models at different levels and in different fields using a suitable assembly strategy;

[0009] Fuse the outputs of models at different levels and in different fields using a suitable fusion mechanism.

[0010] Further, selecting a suitable model according to specific application requirements and problem characteristics specifically includes:

[0011] Select a model from an existing model library or design and develop a new model by oneself;

[0012] When selecting a model, comprehensively consider the model from multiple performance factor perspectives to ensure that the selected model can meet the requirements of practical applications.

[0013] Further, the selected models are divided into different levels, specifically including:

[0014] The models are hierarchically divided by comprehensively considering various performance factors. The divided levels include the underlying model, the middle-level model, and the high-level model;

[0015] The underlying model is the basic model, including geometric models and physical models; the middle-level model is a relatively complex model, including behavioral models; the high-level model is a model with specific functions, including rule models.

[0016] Further, for models in different fields, domain adaptation is performed, specifically including:

[0017] By adopting methods including data preprocessing and feature extraction, models in different fields are converted into a unified format and representation method for assembly and fusion.

[0018] Further, appropriate assembly strategies are used to assemble models of different levels and different fields, specifically including:

[0019] The assembly strategies include serial assembly, parallel assembly, and hybrid assembly methods;

[0020] Serial assembly is to connect models of different levels in sequence to form a series model structure;

[0021] Parallel assembly is to run models of different fields simultaneously and then fuse the results;

[0022] Hybrid assembly is to combine serial assembly and parallel assembly to form a complex model structure.

[0023] Further, appropriate assembly strategies are used to assemble models of different levels and different fields, specifically including:

[0024] Let M1, M2,..., M n be different models. Serial assembly is expressed as: F(x) = M n (M n-1 (...M1(x)...)), where x is the input data and F(x) is the final output. Parallel assembly is expressed as: F(x) = f(M1(x), M2(x),..., M n (x)), where F(·) is the fusion function.

[0025] Further, an appropriate fusion mechanism is used to fuse the outputs of models of different levels and different fields, specifically including:

[0026] The fusion mechanism includes weighted fusion, voting fusion, and deep learning fusion methods.

[0027] Furthermore, an appropriate fusion mechanism is used to fuse the outputs of models at different levels and in different domains, specifically including:

[0028] Weighted fusion assigns a weight to each model according to the performance and importance of different models, and then performs a weighted sum of the outputs of different models to obtain the final output. Specifically, it is expressed as: Let y1, y2,..., y n be the outputs of n models respectively, and w1, w2,..., w n be the weights of n models respectively. Then the output y after weighted fusion is: y = w1y1 + w2y2 +... + w n y n ;

[0029] Voting fusion votes on the outputs of different models and selects the output with the most votes as the final output. Specifically, it is expressed as: Let y1, y2,..., y n be the outputs of n models respectively. For a classification problem, let there be m categories in total. Then the output y after voting fusion is:

[0030]

[0031] where I(·) is the indicator function. When the condition in the parentheses holds, I(·) = 1; otherwise, I(·) = 0;

[0032] Deep learning fusion fuses the outputs of different models through a deep learning model to obtain a more accurate output. Specifically, it is expressed as: Let y1, y2,..., y n be the outputs of n models respectively, and the deep learning fusion model be f(·). Then the output y after deep learning fusion is: y = f(y1, y2,..., y n ).

[0033] According to the second aspect, an embodiment provides a multi-level and multi-domain model assembly and fusion system, which includes:

[0034] A model selection module for selecting a suitable model according to specific application requirements and problem characteristics;

[0035] A hierarchical division module for dividing the selected models into different levels;

[0036] A domain adaptation module for performing domain adaptation on models in different domains;

[0037] An assembly strategy module for assembling models at different levels and in different domains by using a suitable assembly strategy;

[0038] A fusion mechanism module for fusing the outputs of models at different levels and in different fields by using a suitable fusion mechanism.

[0039] According to a third aspect, in one embodiment, an electronic device is provided, and the device includes: a processor and a memory;

[0040] The memory is used for storing one or more program instructions;

[0041] The processor is used for running one or more program instructions to execute the steps of a multi-level and multi-field model assembly and fusion method as described in any one of the above.

[0042] According to a fourth aspect, in one embodiment, a computer-readable storage medium is provided, and a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of a multi-level and multi-field model assembly and fusion method as described in any one of the above are implemented.

[0043] The present application provides a multi-level and multi-field model assembly and fusion method and system, which has the following beneficial effects:

[0044] (1) Improve the performance and generalization ability of the model

[0045] Through the assembly and fusion of multi-level and multi-field models, the advantages of different models can be fully utilized to improve the performance and generalization ability of the model. For example, for image recognition problems, the underlying feature extraction model and the high-level classification model can be assembled and fused to improve the accuracy and robustness of image recognition.

[0046] (2) Enhance the generality and scalability of the model

[0047] The method of the present invention is applicable to models in different fields and of different types, and can be flexibly assembled and fused according to specific application requirements, enhancing the generality and scalability of the model. For example, for natural language processing problems, different language models, part-of-speech tagging models, named entity recognition models, etc. can be assembled and fused to improve the effect of natural language processing.

[0048] (3) Reduce the computing cost and improve the efficiency

[0049] Through reasonable hierarchical division and assembly strategies, the computing cost of the model can be reduced and the efficiency can be improved. For example, for the processing of large-scale data, the underlying models can be distributed for computing, and then the results are fused to improve the computing efficiency. Description of the Drawings

[0050] Figure 1Flowchart of a multi - level and multi - domain model assembly and fusion method provided by an embodiment of the present invention;

[0051] Figure 2 Multi - level model display in a multi - level and multi - domain model assembly and fusion method provided by an embodiment of the present invention;

[0052] Figure 3 Flowchart of model fusion in a multi - level and multi - domain model assembly and fusion method provided by an embodiment of the present invention;

[0053] Figure 4 Schematic diagram of the logical structure of a multi - level and multi - domain model assembly and fusion system provided by an embodiment of the present invention. Detailed implementation manners

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings through specific implementation manners. Similar elements in different implementation manners are labeled with related similar element numbers. In the following implementation manners, many detailed descriptions are provided to enable a better understanding of the present application. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification to avoid overwhelming the core part of the present application with excessive descriptions. For those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the descriptions in the specification and the general technical knowledge in the art.

[0055] In addition, the features, operations, or characteristics described in the specification can be combined in any appropriate manner to form various implementation manners. At the same time, the steps or actions in the method description can also be reordered or adjusted in an obvious manner by those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for clearly describing a certain embodiment and do not mean a necessary sequence unless it is stated that a certain sequence must be followed.

[0056] The first embodiment of the present invention provides a multi - level and multi - domain model assembly and fusion method, which realizes the efficient assembly and fusion of models at different levels and in different domains through innovative technical means, improves the performance and generalization ability of the models, and provides a powerful tool for solving complex problems in multiple domains. The following will be described in detail in conjunction with Figure 1 for detailed description.

[0057] As Figure 1 shown, in step S100, a suitable model is selected according to specific application requirements and problem characteristics.

[0058] Model selection: First, select a suitable model based on the specific application requirements and problem characteristics. You can choose from an existing model library, or you can design and develop a new model yourself. When selecting a model, you need to consider factors such as the model's performance, generalization ability, and computational complexity to ensure that the selected model can meet the needs of the actual application.

[0059] like Figure 1 As shown, in step S200, the selected model is divided into different levels.

[0060] Hierarchical division: Specifically, the selected models are divided into different levels. The models can be divided according to factors such as complexity and function. For example, they can be divided into low-level models, middle-level models and high-level models. Low-level models are usually some basic models, such as geometric models, physical models, etc.; middle-level models are usually some more complex models, such as behavioral models, etc.; high-level models are usually some models with specific functions, such as rule models, etc. Figure 2 shown.

[0061] For the geometric layer, build an extensible geometric metamodel of each equipment part, such as the extensible geometric metamodel of each component such as the lifting column, load-bearing nut, and support head;

[0062] For the physical layer, establish the stress analysis model SAM, fatigue fracture model FCM and other possible material state evolution models for each device during use;

[0063] For the behavior layer, analyze the coupling behaviors and influences of structure, mechanics, etc. between the geometric models of each component, and build a behavior and response model that can describe the behavior sequence, concurrency, linkage and other characteristics of each device;

[0064] For the rule layer, the research constructs rules and logic models for evaluation, optimization, prediction, and traceability that follow and reflect the operation and evolution laws of each device. Then, based on the integrated analysis of "geometry-physics-behavior-rules" and the unified modeling language, the digital twin models of each component of the crane are described in a unified manner.

[0065] like Figure 1 As shown, in step S300, domain adaptation is performed for models in different domains.

[0066] Domain adaptation: Specifically, models in different fields need to be adapted to different fields. Models in different fields can be converted into a unified format and representation through data preprocessing, feature extraction, etc., so as to facilitate assembly and fusion.

[0067] like Figure 1As shown, in step S400, models at different levels and in different domains are assembled using an appropriate assembly strategy.

[0068] Assembly strategy: Specifically, models at different levels and in different domains are assembled using an appropriate assembly strategy. Serial assembly, parallel assembly, hybrid assembly, etc. can be used.

[0069] Serial assembly is to connect models at different levels in sequence to form a series model structure. Parallel assembly is to run models in different domains simultaneously and then fuse the results. Hybrid assembly is to combine serial assembly and parallel assembly to form a more complex model structure.

[0070] Let M1, M2,..., M n be different models. Serial assembly can be expressed as: F(x) = M n (M n-1 (...M1(x)...)), where x is the input data and F(x) is the final output. Parallel assembly can be expressed as: F(x) = f(M1(x), M2(x),..., M n (x)), where F(·) is the fusion function.

[0071] As Figure 1 shown, in step S500, the outputs of models at different levels and in different domains are fused using an appropriate fusion mechanism.

[0072] Fusion mechanism: Specifically, to achieve model fusion, two problems need to be solved first: which models to build and how to build the models. Model fusion considers models other than geometric models, including behavior models and physical models, which can reflect some mechanism characteristics of the modeling object. For a workshop, model fusion should consider models such as information models, process models, and logic models. For equipment, model fusion should consider multi-disciplinary models involved in physical equipment, such as kinematic models and control models. This embodiment studies how to achieve model fusion from a top-level perspective to guide the realization of fusion between models in various domains. Modeling from a top-level perspective is a more abstract description of structure and behavior, which can, to a certain extent, shield the differences between models and contribute to the fusion of multi-disciplinary heterogeneous models.

[0073] The outputs of models at different levels and in different domains are fused using an appropriate fusion mechanism. Weighted fusion, voting fusion, deep learning fusion, etc. can be used, as follows:

[0074] a. Weighted fusion is to assign a weight to each model according to the performance and importance of different models, and then perform weighted summation of the outputs of different models to obtain the final output. Let y1, y2,..., yn are the outputs of n models respectively, and w1, w2, ..., w n are the weights of n models respectively. Then the output y after weighted fusion is: y = w1y1 + w2y2 +... + w n y n .

[0075] b. Voting fusion is to vote on the outputs of different models and select the output with the most votes as the final output. Let y1, y2, ..., y n be the outputs of n models respectively. For a classification problem, assume there are m categories in total. Then the output y after voting fusion is:

[0076]

[0077] where I(·) is the indicator function. When the condition in the parentheses holds, I(·) = 1; otherwise, I(·) = 0.

[0078] c. Deep learning fusion is to fuse the outputs of different models through a deep learning model to obtain a more accurate output. Let y1, y2, ..., y n be the outputs of n models respectively. The deep learning fusion model is f(·). Then the output y after deep learning fusion is: y = f(y1, y2, ..., y n ).

[0079] Similar to the assembly of models, models related to mechanisms also have hierarchical characteristics. Although there are some differences in the models to be fused, from the perspective of top-level modeling, the process of model fusion is the same. Therefore, this embodiment proposes a general model fusion process. As Figure 3 shown, the process of model fusion mainly includes five steps, and the specific content of each step is as follows.

[0080] (1) Hierarchical division. The top-level model is an abstraction of models in each field. Therefore, the hierarchical division of the top-level model should be defined according to the characteristics of the modeling object. In addition, the hierarchy of models in each field should be as consistent as possible with the hierarchy in model assembly, which helps to establish the mapping relationship between geometric models and physical models or behavior models.

[0081] (2) Representation of models in each field. After constructing the models in each field of the modeling object, these models are abstracted into a descriptive model, that is, the parameters, attribute values, input interfaces, and output interfaces of the models in each field need to be defined in this step.

[0082] (3) Analysis of coupling mechanism. Based on the descriptive model constructed in step (2), the coupling mechanism should be analyzed from the perspectives of the workshop and equipment respectively. In addition, these relationships should be represented by the connections or constraints between the abstract descriptive models to guide the model fusion among models in various fields.

[0083] (4) Implementation of model fusion. After defining the connections and constraints between different models, some modeling languages and corresponding software can be used to implement model fusion. Currently, there have been some research on the conversion from descriptive models (such as the system modeling language SysML) to simulation models (such as Simulink).

[0084] (5) Model update. Different from the model update of geometric models, the fusion model needs to consider not only the update of model parameters but also the update of model relationships, such as the update of process models. According to the physical and behavioral characteristics of the modeling object, the update parameters or logical relationships and the update frequency can be defined. In addition, the update of the model can be supported by constructing the mapping relationship between the collected data and model parameters.

[0085] In this embodiment, the fault diagnosis of the fixed car lifting machine equipment is taken as an example to introduce the specific implementation method.

[0086] (I) Model selection

[0087] 1. Select the underlying sensor data acquisition models, such as current sensor, voltage sensor, displacement sensor models, etc., to collect the operation data of the fixed car lifting machine equipment.

[0088] 2. Select the middle-level feature extraction models, such as the principal component analysis (PCA) model, independent component analysis (ICA) model, etc., to extract the features of the sensor data.

[0089] 3. Select the high-level fault diagnosis models, such as the support vector machine (SVM) model, artificial neural network (ANN) model, etc., to diagnose the faults of the fixed car lifting machine equipment.

[0090] (II) Hierarchical division

[0091] 1. Take the sensor data acquisition model as the underlying model to collect the operation data of the fixed car lifting machine equipment, such as current, voltage, displacement, etc.

[0092] 2. Take the feature extraction model as the middle-level model to extract the features of the sensor data, such as mean value, variance, frequency, etc.

[0093] 3. Take the fault diagnosis model as the high-level model to diagnose the faults of the fixed car lifting machine equipment, such as judging whether there are motor faults, mechanical faults, etc.

[0094] (3) Domain Adaptation

[0095] 1. For sensor data, perform preprocessing such as filtering and noise reduction, and convert it into a format suitable for input to the sensor data acquisition model.

[0096] 2. For the feature extraction model and the fault diagnosis model, convert the data collected by the sensor data acquisition model to meet the input requirements of the feature extraction model and the fault diagnosis model.

[0097] (4) Assembly Strategy

[0098] 1. Adopt a serial assembly method to connect the sensor data acquisition model, the feature extraction model, and the fault diagnosis model in sequence.

[0099] 2. First, input the operation data of the fixed car lifter equipment into the sensor data acquisition model to collect relevant data. Then, input the collected data into the feature extraction model to extract the features of the data. Finally, input the extracted features into the fault diagnosis model for fault diagnosis.

[0100] (5) Fusion Mechanism

[0101] 1. Adopt a weighted fusion method to assign different weights to the sensor data acquisition model, the feature extraction model, and the fault diagnosis model.

[0102] 2. Determine the weight values through experiments according to the performance and importance of the models. Then, perform weighted summation on the outputs of different models to obtain the final fault diagnosis result.

[0103] 3. Assume that the outputs of the sensor data acquisition model, the feature extraction model, and the fault diagnosis model are y1, y2, and y3 respectively, and their weights are w1, w2, and w3 respectively. Then the final output is y = w1y1 + w2y2 + w3y3.

[0104] Corresponding to the above-disclosed method for assembling and fusing a multi-level and multi-domain model, an embodiment of the present invention also discloses a multi-level and multi-domain model assembling and fusing system, as Figure 4 shown, which specifically includes:

[0105] A model selection module, configured to select a suitable model according to specific application requirements and problem characteristics;

[0106] A hierarchical division module, configured to divide the selected model into different levels;

[0107] A domain adaptation module, configured to perform domain adaptation on models in different domains;

[0108] An assembly strategy module for assembling models at different levels and in different domains using appropriate assembly strategies;

[0109] A fusion mechanism module for fusing the outputs of models at different levels and in different domains using appropriate fusion mechanisms.

[0110] It should be noted that for a detailed description of a multi-level and multi-domain model assembly and fusion system provided in an embodiment of the present invention, reference can be made to the relevant description of a multi-level and multi-domain model assembly and fusion method provided in an embodiment of the present application, which will not be elaborated here.

[0111] In addition, an embodiment of the present invention also provides an electronic device, which includes: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a multi-level and multi-domain model assembly and fusion method as described in any one of the above.

[0112] It should be noted that for a detailed description of an electronic device provided in an embodiment of the present invention, reference can be made to the relevant description of a multi-level and multi-domain model assembly and fusion method provided in an embodiment of the present application, which will not be elaborated here.

[0113] In addition, an embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of a multi-level and multi-domain model assembly and fusion method as described in any one of the above.

[0114] It should be noted that for a detailed description of a computer-readable storage medium provided in an embodiment of the present invention, reference can be made to the relevant description of a multi-level and multi-domain model assembly and fusion method provided in an embodiment of the present application, which will not be elaborated here.

[0115] Those skilled in the art can understand that all or part of the functions of the various methods in the above embodiments can be implemented in a hardware manner or in a computer program manner. When all or part of the functions in the above embodiments are implemented in a computer program manner, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc. The above functions can be realized by the computer executing the program. For example, the program is stored in the memory of the device, and when the processor executes the program in the memory, all or part of the above functions can be realized. In addition, when all or part of the functions in the above embodiments are implemented in a computer program manner, the program can also be stored in a storage medium such as a server, another computer, magnetic disk, optical disk, flash drive or mobile hard disk, and saved to the memory of the local device by downloading or copying, or the system of the local device is updated. When the processor executes the program in the memory, all or part of the functions in the above embodiments can be realized.

[0116] The above uses specific examples to elaborate on the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art of the present invention, according to the idea of the present invention, several simple deductions, deformations or substitutions can also be made.

Claims

1. A multi-level, multi-domain model assembly and fusion method, characterized in that: The method comprises: Select the appropriate model based on specific application requirements and problem characteristics; Divide the selected models into different levels; For models in different fields, perform domain adaptation; Use appropriate assembly strategies to assemble models at different levels and in different fields; Use appropriate fusion mechanisms to fuse the outputs of models at different levels and in different fields.

2. A multi-level, multi-domain model assembly and fusion method as claimed in claim 1, characterized in that: Select the appropriate model based on specific application requirements and problem characteristics, including: Select models from existing model libraries or design and develop new models on your own; When selecting a model, the model is comprehensively considered from the perspective of multiple performance factors to ensure that the selected model can meet the needs of actual applications.

3. A multi-level, multi-domain model assembly and fusion method as claimed in claim 1, characterized in that: The selected models are divided into different levels, including: Taking into account a variety of performance factors, the model is divided into layers, including the bottom-level model, the middle-level model and the high-level model; The bottom-level model is a basic model, including a geometric model and a physical model; the middle-level model is a relatively complex model, including a behavioral model; and the high-level model is a model with specific functions, including a rule model.

4. A multi-level, multi-domain model assembly and fusion method as claimed in claim 1, characterized in that: For models in different fields, domain adaptation is performed, including: By adopting methods including data preprocessing and feature extraction, models from different fields are converted into a unified format and representation for assembly and fusion.

5. A multi-level, multi-domain model assembly and fusion method as claimed in claim 1, characterized in that: Use appropriate assembly strategies to assemble models at different levels and in different fields, including: Assembly strategies include serial assembly, parallel assembly, and hybrid assembly; Serial assembly is to connect models at different levels in sequence to form a serial model structure; Parallel assembly is to run models from different fields simultaneously and then fuse the results; Hybrid assembly combines serial assembly and parallel assembly to form a complex model structure.

6. A multi-level, multi-domain model assembly and fusion method as claimed in claim 5, characterized in that: Use appropriate assembly strategies to assemble models at different levels and in different fields, including: Let M1, M1, ..., M n For different models, serial assembly is expressed as: F(x) = M n (M n-1 (...M1(x)...)), where x is the input data and F(x) is the final output. The parallel assembly is expressed as: F(x) = f(M1(x), M2(x), ..., M n (x)), where F(·) is the fusion function.

7. A multi-level, multi-domain model assembly and fusion method as claimed in claim 1, characterized in that: Use appropriate fusion mechanisms to fuse the outputs of models at different levels and in different fields, including: The fusion mechanisms include weighted fusion, voting fusion and deep learning fusion.

8. A multi-level, multi-domain model assembly and fusion method as claimed in claim 7, characterized in that: Use appropriate fusion mechanisms to fuse the outputs of models at different levels and in different fields, including: Weighted fusion is to assign a weight to each model according to the performance and importance of different models, and then perform weighted summation of the outputs of different models to obtain the final output. It is specifically expressed as: Let y1, y2, ..., y n are the outputs of n models, w1, w1, ..., w n are the weights of n models respectively, then the output y after weighted fusion is: y=w1y1+w2y2+...+w n y n ; Voting fusion is to vote on the outputs of different models and select the output with the most votes as the final output. The specific expression is: let y1, y2, ..., y n are the outputs of n models respectively. For a classification problem, assuming there are m categories in total, the output y after voting fusion is: Where I(·) is an indicator function. When the condition in the brackets is met, I(·) = I, otherwise I(·) = 0; Deep learning fusion is to fuse the outputs of different models through deep learning models to obtain more accurate outputs. Specifically, let y1, y2, ..., y n are the outputs of n models respectively, and the deep learning fusion model is f(·). Then the output y after deep learning fusion is: y=f(y1,y2,...,y n ).

9. A multi-level, multi-domain model assembly and fusion system, characterized in that: The system comprises: Model selection module, used to select the appropriate model according to specific application requirements and problem characteristics; The hierarchical division module is used to divide the selected model into different hierarchies; The domain adaptation module is used to adapt models in different fields. The assembly strategy module is used to assemble models of different levels and fields using appropriate assembly strategies; The fusion mechanism module is used to fuse the outputs of models at different levels and in different fields using a suitable fusion mechanism.

10. An electronic device, characterized in that: The device comprises: a processor and a memory; The memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a multi-level, multi-domain model assembly and fusion method as described in any one of claims 1 to 8.

Images