Multi-model fusion processing and automatic switching method and system

Abstract

The invention relates to the technical field of model processing, and particularly discloses a multi-model fusion processing and automatic switching method. Comprising the steps of obtaining input data of a user, performing data preprocessing on the input data, inputting the processed data into a plurality of different types of pre-training models, performing statistics on output results of all the models, performing fusion processing on the collected results, and obtaining a fusion result; the method comprises the steps that firstly, a deep learning model, a traditional machine learning model and a statistical model are integrated, then performance parameters of all the models are monitored in real time based on performance monitoring indexes, and the current model is automatically switched according to the performance parameters of the real-time monitored models. The advantages of the models in different data types and task scenes can be fully exerted, image feature extraction can be processed through the deep learning model, rule mining of structured numerical data can be coped with through the traditional machine learning model, data distribution correlation analysis can be carried out through the statistical model, and the accuracy of industrial detection results can be improved.

Description

Technical Field

[0001] The present invention relates to the field of model processing technology, and in particular to a method and system for multi-model fusion processing and automatic switching. Background Art

[0002] In today's complex data processing and analysis scenarios, a single model often struggles to cope with diverse task requirements and data characteristics. With the explosive growth of data volumes and the increasing diversity of data types, from structured to unstructured data, different models have demonstrated their respective strengths and limitations.

[0003] Deep learning models, such as convolutional neural networks (CNNs), can automatically extract complex image features in image recognition tasks and possess powerful representational learning capabilities. However, they can be prone to overfitting on small sample sizes and exhibit poor interpretability. Traditional machine learning models, such as decision tree algorithms, excel when processing small-scale data with high interpretability requirements, with clear and easy-to-understand decision rules. However, they can face challenges with high computational complexity and limited accuracy when working with large-scale, high-dimensional data. Statistical models have unique applications in data modeling tasks based on probability distribution assumptions. However, their strict assumptions about data distribution can reduce their applicability in complex real-world data environments.

[0004] In practical applications, a single model cannot fully utilize the information of various types of data. For example, fault detection in industrial production is prone to false detection or missed detection because it is necessary to comprehensively consider multiple sources of data such as equipment operation data and environmental data. A single model is difficult to capture the complex relationships and potential patterns between data, and cannot detect faults in a timely and accurate manner.

[0005] Therefore, there is an urgent need for a multi-model fusion processing and automatic switching method and system to adapt to complex and changeable data processing and analysis tasks and improve the accuracy, stability and adaptability of detection. Summary of the Invention

[0006] The object of the present invention is to provide a method for multi-model fusion processing and automatic switching, comprising: Obtain industrial inspection data input by users and perform data preprocessing on the input data, where data preprocessing includes data normalization and feature extraction; Inputting the processed data into a plurality of different types of pre-trained models, including deep learning models, traditional machine learning models, and statistical models; Calculate statistics on the output results of each model; Performing fusion processing on the collected results, wherein the fusion processing includes weighted fusion; Setting performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current detection task; Real-time monitoring of the performance parameters of each model based on performance monitoring indicators; If the performance parameters of the current model are lower than the set performance threshold, the current model will be automatically switched according to the performance parameters of the real-time monitoring model to continue the processing of industrial detection data and obtain the detection results.

[0007] Furthermore, the step of data normalization includes: Obtaining the maximum and minimum values within a sampling period according to the input data, and calculating the data range according to the maximum and minimum values; Performing a linear transformation on the input data according to the data range and a preset normalization range to obtain normalized data; Perform time-frequency conversion on the normalized data to convert the time domain signal into the frequency domain signal; Divide the frequency domain signal into multiple equal-width frequency bands according to a preset frequency band division rule; Performing energy integration on the spectrum within each of the equal-width frequency bands to extract energy distribution characteristics of each frequency band; A multidimensional feature vector is constructed according to the energy distribution characteristics, and the multidimensional feature vector is used as standardized data for subsequent model input.

[0008] Furthermore, the collected results are fused, wherein the fusion process includes weighted fusion, and the weighted fusion step includes: According to the preset model weight coefficients, the output results of the deep learning model, traditional machine learning model and statistical model are weighted respectively, where the sum of the weight coefficients of all models is equal to 1; Establish a weighted fusion calculation rule and multiply the output of each model with its corresponding weight coefficient; Linearly superimpose the product results of each model to generate the final output result after fusion processing; A weight normalization constraint condition is set so that the weight coefficients always maintain a unit sum relationship during the model switching process.

[0009] Furthermore, the step of automatically switching the current model includes: Pre-store backup models; Determine the performance indicators of the currently running model; If the performance index of the current model is higher than the preset performance threshold, the current model will continue to run; If the performance index of the current model is lower than the preset performance threshold, the best performing model is selected from the backup models according to the performance ranking to replace it.

[0010] Furthermore, the preset performance threshold range is 80%. , is the accuracy of the current model in the optimal state, that is, when the performance index of real-time monitoring is lower than 80% , the current model is switched.

[0011] Furthermore, the step of automatically switching the current model also includes: Recording model switching and generating a switching record database, wherein the switching record database includes the switching time, the original model, the new model and the switching reason; After switching the model, evaluate the performance of the new model to ensure that the performance of the new model meets expectations. If it does not meet expectations, switch again or fine-tune the new model.

[0012] Furthermore, the present invention also discloses a system for multi-model fusion processing and automatic switching, comprising: The acquisition module is used to obtain the user's input data and perform data preprocessing on the input data, wherein the data preprocessing includes data normalization and feature extraction; An input module, configured to input the processed data into a plurality of different types of pre-trained models, including deep learning models, traditional machine learning models, and statistical models; Statistics module, used to collect statistics on the output results of each model; A fusion module is used to perform fusion processing on the collected results, wherein the fusion processing includes weighted fusion; A preset module is used to set performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current task; The monitoring module is used to monitor the performance parameters of each model in real time based on performance monitoring indicators; The switching module is used to automatically switch the current model according to the performance parameters of the real-time monitoring model when the performance parameters of the current model are lower than the set performance threshold.

[0013] Furthermore, the switching module includes: A reserve unit, used for pre-storing spare models; A determination unit, used to determine the performance indicators of the currently running model; A maintenance unit, configured to continue running the current model when the performance indicator of the current model is higher than a preset performance threshold; The switching unit is used to select the best-performing model from the backup models according to the performance ranking to replace the current model when the performance index of the current model is lower than the preset performance threshold.

[0014] The present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0015] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0016] The beneficial effects of this application are: By integrating deep learning models, traditional machine learning models, and statistical models, this invention can fully leverage the advantages of each model across different data types and task scenarios, overcoming the limitations of a single model. Deep learning models can handle image feature extraction, traditional machine learning models can handle rule mining for structured numerical data, and statistical models can perform data distribution correlation analysis. Together, these approaches can handle tasks more comprehensively and accurately, improving the accuracy and reliability of industrial inspection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a method flow chart proposed in one embodiment of the present application; The realization of the objectives, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] like Figure 1 As shown, the present application provides a method for multi-model fusion processing and automatic switching, including: S1, obtains industrial inspection data input by the user and performs data preprocessing on the input data, where the data preprocessing includes data normalization and feature extraction; S2, inputting the processed data into a plurality of different types of pre-trained models, wherein the pre-trained models include deep learning models, traditional machine learning models, and statistical models; S3, statistics the output results of each model; S4, performing fusion processing on the collected results, wherein the fusion processing includes weighted fusion; S5, setting performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current detection task; S6, real-time monitoring of the performance parameters of each model based on performance monitoring indicators; S7, if the performance parameters of the current model are lower than the set performance threshold, the current model is automatically switched according to the performance parameters of the real-time monitoring model to continue the processing of the industrial detection data and obtain the detection results.

[0020] As described in steps S1-S7 above, the present invention can fully utilize the advantages of each model in different data types and task scenarios by integrating deep learning models, traditional machine learning models, and statistical models, overcoming the limitations of a single model. Deep learning models can handle image feature extraction, traditional machine learning models can handle rule mining of structured numerical data, and statistical models can perform data distribution correlation analysis. Combined, they can handle tasks more comprehensively and accurately, improving the accuracy and reliability of industrial inspection results.

[0021] Based on performance monitoring indicators and automatic switching mechanisms, it is also possible to perceive changes in model performance in real time. When data characteristics or task requirements change, such as data distribution offsets or task focus shifts, once the performance of a certain model falls below the threshold, it can be promptly switched to a more suitable backup model to ensure that the system is always in a better performance state. By automatically completing performance monitoring and switching, there is no need for engineers to constantly monitor or manually handle performance degradation after it is discovered, reducing the operational burden and the risk of response delays. Through the above effects, especially improving accuracy, reducing missed detections, and ensuring continuity, this solution ultimately effectively guarantees product quality, equipment safety, and the stability of the production process, and reduces the potential economic losses and safety accident risks caused by detection errors.

[0022] Specifically, the data normalization step includes: Obtaining the maximum and minimum values within a sampling period according to the input data, and calculating the data range according to the maximum and minimum values; Performing a linear transformation on the input data according to the data range and a preset normalization range to obtain normalized data; Perform time-frequency conversion on the normalized data to convert the time domain signal into the frequency domain signal; Divide the frequency domain signal into multiple equal-width frequency bands according to a preset frequency band division rule; Performing energy integration on the spectrum within each of the equal-width frequency bands to extract energy distribution characteristics of each frequency band; A multidimensional feature vector is constructed according to the energy distribution characteristics, and the multidimensional feature vector is used as standardized data for subsequent model input.

[0023] S8, the specific normalization process is: for the input data, normalize it to the interval [0,1] or [-1,1], and the normalization formula used is: When normalized to the interval [0,1], the formula is: ; When normalized to the interval [-1,1], the formula is: ; Here, α is the eigenvalue.

[0024] As described in step S8 above, in practical applications, input data often comes from diverse sources, and their numerical ranges can vary widely. For example, in a health risk assessment task that combines medical test indicators (values typically fall within a relatively small, reasonable range) with basic patient information (such as age, which has a relatively wide numerical range), normalizing the data to the interval [0, 1] or [-1, 1] can unify these originally scaled data into the same standard range. This prevents bias in some models during training or testing due to large differences in numerical magnitude, ensuring that different types of data are treated equally in subsequent model processing.

[0025] For the model used in this method, data normalization helps optimize the model training process. For optimization algorithms based on gradient descent, normalized data can smooth the model's loss function, accelerate model convergence, reduce training time, and, to a certain extent, reduce the risk of the model falling into a local optimal solution, thereby improving model training efficiency and ultimate performance.

[0026] Specifically, the collected results are fused, wherein the fusion process includes weighted fusion, and the weighted fusion step includes: According to the preset model weight coefficients, the output results of the deep learning model, traditional machine learning model and statistical model are weighted respectively, where the sum of the weight coefficients of all models is equal to 1; Establish a weighted fusion calculation rule and multiply the output of each model with its corresponding weight coefficient; Linearly superimpose the product results of each model to generate the final output result after fusion processing; A weight normalization constraint condition is set so that the weight coefficients always maintain a unit sum relationship during the model switching process.

[0027] S9, performing weighted summation on the output results of each model according to a pre-set weight; The summation formula is: ; in is the output of the i-th model, is the weight of the i-th model, and .

[0028] As described in step S9 above, the output results of multiple different types of models can be fused through weighted summation. Each model may have its own advantages and disadvantages in different data types or task scenarios. This fusion method can fully utilize the advantages of each model to obtain more comprehensive and accurate results. For example, in a comprehensive detection task with both image data and text data, a deep learning model may perform well in processing image data, while a traditional machine learning model has its unique advantages in processing text data. Weighted summation can combine the advantages of both.

[0029] Weight It is pre-set, which means that the weights of each model can be flexibly adjusted according to different task requirements, data characteristics, and the performance of the model in a specific scenario. If, in a certain task, it is found through evaluation that the deep learning model is more reliable, its weight can be appropriately increased so that the final result is more biased towards the output of this model. The formula stipulates that the sum of all model weights is 1. This constraint ensures that the result of the weighted summation is within a reasonable range, avoiding excessive or insufficient deviations in the results due to improper weight setting, making the final fusion result more stable and reliable. And because the sum of the weights is 1, the result obtained by weighted summation has a clear meaning and comparability. In actual detection tasks, such results are easier to understand and interpret, and are also convenient for comparison with other methods or benchmark results.

[0030] Specifically, the step of automatically switching the current model includes: S10, pre-storing a backup model; S11, determining the performance indicators of the currently running model; S12, if the performance index of the current model is higher than the preset performance threshold, continue to run the current model; S13: If the performance indicator of the current model is lower than a preset performance threshold, a model with the best performance is selected from the backup models according to the performance ranking to replace it.

[0031] As described in steps S10-S13 above, by pre-storing backup models, the backup model can be quickly activated in the event of unexpected situations such as data attacks, extreme data points affecting the performance of the current model, or unexpected failures in the current model, avoiding system interruptions and ensuring continuous and stable system operation. In the field of industrial inspection, where the distribution and characteristics of data change over time, an automatic switching mechanism can be used to promptly replace the existing model with a more suitable backup model when the performance of the existing model drops below a preset threshold due to data changes. This ensures that the system can always effectively process data, safeguards the continuity of the inspection process, and minimizes the impact of single points of failure. For example, in high-speed, continuous industrial production lines (such as automotive manufacturing, electronics assembly, and food packaging), inspection interruptions or delays often mean the risk of batch product release or production line stagnation. By pre-storing backup models and establishing a model resource pool, when the performance of the primary model is detected to be degraded (performance indicators fall below a threshold), the optimal backup model is automatically selected and switched to instantly. This switching mechanism ensures that the core processing unit of the detection task will not pause due to the failure of a single model, effectively avoiding the interruption of the detection process caused by model failure and preventing the possible missed detection problems.

[0032] In addition, it can dynamically maintain high-precision detection and significantly reduce the risk of missed detection. For example, in industrial scenarios (such as equipment vibration monitoring, surface defect detection, and weld quality assessment), data distribution can drift due to equipment wear, environmental disturbances, material batch differences, or the emergence of unprecedented abnormal patterns. Through the closed-loop monitoring and switching system, the performance degradation of the main model can be sensed in real time. Once its key indicators such as accuracy and recall rate (especially those sensitive to missed detection) fall below the preset safety threshold (such as 80%), the best backup model can be immediately activated to replace it. This ensures that at any time, the core model driving the detection decision is in an acceptable high-performance state, continuously maintaining a high detection rate in dynamic changes, and especially reducing the risk of systematic missed detection.

[0033] It should be noted that the present invention can also avoid unnecessary resource consumption. Switching is only performed when the performance indicators of the current model are indeed lower than the preset threshold, avoiding the waste of computing resources due to slight performance fluctuations or unnecessary model replacement. For systems that process large-scale data and complex models, this helps to improve resource utilization efficiency and reduce operating costs. The entire switching process is carried out automatically based on pre-set rules and does not require manual continuous monitoring and frequent intervention. This can significantly reduce labor costs in long-term running systems, while also reducing the errors and uncertainties that may be caused by manual operations.

[0034] Specifically, in S14, the preset performance threshold range is 80%. , is the accuracy of the current model in the optimal state, that is, when the performance index of real-time monitoring is lower than 80% , the current model is switched.

[0035] As described in step S14 above, by setting a preset performance threshold, excessive performance degradation can be prevented. Once the performance indicator falls below the threshold, switching is performed to prevent excessive degradation of system performance. In the field of industrial detection, using models with poor performance can have serious consequences. This mechanism can replace poorly performing models in a timely manner to ensure the stability and security of the system. For users, the system can continuously maintain good performance and provide accurate and reliable results, thereby improving the user experience. Whether it is an online service system or an application in a smart device, a stable and high-performance model can better meet the needs of users.

[0036] Specifically, the step of automatically switching the current model further includes: S15, recording the model switching and generating a switching record database, wherein the switching record database includes the switching time, the original model, the new model and the switching reason; S16, after switching the model, the performance of the new model is evaluated to ensure that the performance of the new model meets expectations. If it does not meet expectations, switch again or fine-tune the new model.

[0037] As described in steps S15-S16 above, by recording the time, original model, new model, and reason for the model switch, the entire model switch process can be traced. In a complex data processing and analysis system, when a problem occurs or when it is necessary to review system performance changes, the relevant information of each model switch can be easily queried, which helps to quickly locate the problem. The switch record database can provide data support for system optimization. By analyzing these records, it can be found which models are prone to performance degradation under specific circumstances, thereby triggering a switch, which helps R&D personnel to improve the model itself or the model switching strategy and improve the overall performance of the system.

[0038] For industries with high safety requirements, such as automotive manufacturing, electronics assembly, and food packaging, a complete model switching record helps meet regulatory requirements. It serves as the basis for transparent system operations, ensuring that every step of the system's operation is based on evidence and auditable. After switching the model, a performance evaluation of the new model is conducted to ensure that its performance meets expectations. Performance evaluation of the newly switched model can promptly determine whether the new model can meet the current mission requirements of the system. If the new model's performance does not meet expectations, immediate measures can be taken, such as switching again or fine-tuning, to ensure that the system can continue to operate stably after the model switch and avoid system performance degradation caused by an inappropriate model.

[0039] Specifically, the present invention also discloses a system for multi-model fusion processing and automatic switching, comprising: The acquisition module is used to obtain the user's input data and perform data preprocessing on the input data, wherein the data preprocessing includes data normalization and feature extraction; An input module, configured to input the processed data into a plurality of different types of pre-trained models, including deep learning models, traditional machine learning models, and statistical models; Statistics module, used to collect statistics on the output results of each model; A fusion module is used to perform fusion processing on the collected results, wherein the fusion processing includes weighted fusion; A preset module is used to set performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current task; The monitoring module is used to monitor the performance parameters of each model in real time based on performance monitoring indicators; The switching module is used to automatically switch the current model according to the performance parameters of the real-time monitoring model when the performance parameters of the current model are lower than the set performance threshold.

[0040] Specifically, the switching module includes: A reserve unit, used for pre-storing spare models; A determination unit, used to determine the performance indicators of the currently running model; A maintenance unit, configured to continue running the current model when the performance indicator of the current model is higher than a preset performance threshold; The switching unit is used to select the best-performing model from the backup models according to the performance ranking to replace the current model when the performance index of the current model is lower than the preset performance threshold.

[0041] The present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0042] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0043] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, value library or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and RAM bus dynamic RAM (RDRAM).

[0044] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0045] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent results or equivalent process transformations made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for multi-model fusion processing and automatic switching, characterized in that: include: Obtain industrial inspection data input by users and perform data preprocessing on the input data, where data preprocessing includes data normalization and feature extraction; Inputting the processed data into a plurality of different types of pre-trained models, including deep learning models, traditional machine learning models, and statistical models; Calculate statistics on the output results of each model; Performing fusion processing on the collected results, wherein the fusion processing includes weighted fusion; Setting performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current detection task; Real-time monitoring of the performance parameters of each model based on performance monitoring indicators; If the performance parameters of the current model are lower than the set performance threshold, the current model will be automatically switched according to the performance parameters of the real-time monitoring model to continue the processing of industrial detection data and obtain the detection results.

2. The method for multi-model fusion processing and automatic switching according to claim 1, characterized in that: The data normalization step includes: Obtaining the maximum and minimum values within a sampling period according to the input data, and calculating the data range according to the maximum and minimum values; Performing a linear transformation on the input data according to the data range and a preset normalization range to obtain normalized data; Perform time-frequency conversion on the normalized data to convert the time domain signal into the frequency domain signal; Divide the frequency domain signal into multiple equal-width frequency bands according to a preset frequency band division rule; Performing energy integration on the spectrum within each of the equal-width frequency bands to extract energy distribution characteristics of each frequency band; A multidimensional feature vector is constructed according to the energy distribution characteristics, and the multidimensional feature vector is used as standardized data for subsequent model input.

3. The method for multi-model fusion processing and automatic switching according to claim 2, characterized in that: The collected results are fused, wherein the fusion process includes weighted fusion, and the weighted fusion step includes: According to the preset model weight coefficients, the output results of the deep learning model, traditional machine learning model and statistical model are weighted respectively, where the sum of the weight coefficients of all models is equal to 1; Establish a weighted fusion calculation rule and multiply the output of each model with its corresponding weight coefficient; Linearly superimpose the product results of each model to generate the final output result after fusion processing; A weight normalization constraint condition is set so that the weight coefficients always maintain a unit sum relationship during the model switching process.

4. The method for multi-model fusion processing and automatic switching according to claim 1, characterized in that: The step of automatically switching the current model includes: Pre-store backup models; Determine the performance indicators of the currently running model; If the performance index of the current model is higher than the preset performance threshold, the current model will continue to run; If the performance index of the current model is lower than the preset performance threshold, the best performing model is selected from the backup models according to the performance ranking to replace it.

5. The method for multi-model fusion processing and automatic switching according to claim 4, characterized in that: The preset performance threshold range is 80% , is the accuracy of the current model in the optimal state, that is, when the performance index of real-time monitoring is lower than 80% , the current model is switched.

6. The method for multi-model fusion processing and automatic switching according to claim 5, characterized in that: The step of automatically switching the current model further includes: Recording model switching and generating a switching record database, wherein the switching record database includes the switching time, the original model, the new model and the switching reason; After switching the model, evaluate the performance of the new model to ensure that the performance of the new model meets expectations. If it does not meet expectations, switch again or fine-tune the new model.

7. A system for multi-model fusion processing and automatic switching, characterized in that: include: The acquisition module is used to obtain the user's input data and perform data preprocessing on the input data, wherein the data preprocessing includes data normalization and feature extraction; An input module, configured to input the processed data into a plurality of different types of pre-trained models, including deep learning models, traditional machine learning models, and statistical models; Statistics module, used to collect statistics on the output results of each model; A fusion module is used to perform fusion processing on the collected results, wherein the fusion processing includes weighted fusion; A preset module is used to set performance monitoring indicators, which are used to evaluate the performance parameters of each model in the current task; The monitoring module is used to monitor the performance parameters of each model in real time based on performance monitoring indicators; The switching module is used to automatically switch the current model according to the performance parameters of the real-time monitoring model when the performance parameters of the current model are lower than the set performance threshold.

8. The multi-model fusion processing and automatic switching system according to claim 7, characterized in that: The switching module includes: A reserve unit, used for pre-storing spare models; A determination unit, used to determine the performance indicators of the currently running model; A maintenance unit, configured to continue running the current model when the performance indicator of the current model is higher than a preset performance threshold; The switching unit is used to select the best-performing model from the backup models according to the performance ranking to replace the current model when the performance index of the current model is lower than the preset performance threshold.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for multi-model fusion processing and automatic switching described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for multi-model fusion processing and automatic switching described in any one of claims 1 to 6 are implemented.

Images