AI cluster training method adopting distributed energy and related device

Abstract

The invention provides an AI cluster training method adopting distributed energy, belongs to the technical field of artificial intelligence, and can effectively utilize the distributed energy and improve the efficiency and accuracy of AI cluster training. The computing tasks are reasonably distributed among the local equipment, the edge side equipment and the cloud equipment, so that computing resources of the equipment can be fully utilized, the data transmission time is shortened, and the delay is reduced. In addition, by adopting an image segmentation algorithm and training of a deep learning model, the precision and speed of data processing can be further improved. Finally, by integrating processing results of different devices, a more comprehensive and accurate reasoning result can be obtained, and application requirements in a complex scene are met.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to an AI cluster training method and related devices using distributed energy. Background Art

[0002] In today's era of rapid technological development, artificial intelligence (AI) technology has been widely used in various fields, from self-driving cars to intelligent voice assistants to medical diagnosis systems. However, the rapid development of AI technology is inseparable from large-scale computing resources and efficient data processing capabilities. Traditional centralized data centers can no longer meet the high computing requirements of AI training. Therefore, the AI ​​cluster training method using distributed energy has emerged.

[0003] In addition, distributed energy systems can improve the flexibility and scalability of AI clusters. Traditional centralized data centers usually require a lot of infrastructure investment, while distributed energy systems can flexibly expand computing resources through modular design. In this way, computing nodes can be quickly increased or decreased according to actual needs, improving resource utilization and reducing operating costs.

[0004] During the AI ​​cluster training process, the distributed energy system can also achieve the optimal match between computing resources and energy through intelligent scheduling algorithms. For example, according to the real-time needs of the training task, the energy supply of the computing node is dynamically adjusted to ensure the efficient use of computing resources. At the same time, by real-time monitoring and management of energy consumption, energy efficiency can be further optimized and energy waste can be reduced. Generally, a centralized training room is used, and model training is performed on manually labeled samples. The process is cumbersome and inefficient. The edge device is only responsible for reasoning applications based on the model and cannot participate in the model training process. Summary of the invention

[0005] The embodiments of the present application provide an AI cluster training method and related devices using distributed energy to improve the above-mentioned problems.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions:

[0007] In a first aspect, an embodiment of the present application proposes an AI cluster training method using distributed energy, which is applicable to an AI cluster training system using distributed energy. The system includes a control terminal, an edge device, a local device, and a cloud device. The method includes:

[0008] The control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling;

[0009] The control terminal allocates target inference samples based on the predicted total computing power requirements and meets the following requirements:

[0010] When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target inference sample so that the target inference sample forms a first sub-sample and a second sub-sample;

[0011] The control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing;

[0012] The control terminal obtains the processing results of the cloud device and integrates the processing results with the processing results of the local device and the edge device.

[0013] In combination with the first aspect, in some implementations, the control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on the result of the preliminary sampling, including:

[0014] The control terminal obtains a target reasoning sample, obtains a prediction sample from the target reasoning sample based on a random sampling method, and obtains a corresponding relationship between the prediction sample and the target reasoning sample;

[0015] The control terminal imports the prediction sample into the training device, and obtains the unit computing power requirement corresponding to the prediction sample based on the training result and training time fed back by the training device;

[0016] The control terminal determines the predicted total computing power requirement of the target sample based on the correspondence between the predicted sample and the target inference sample, as well as the unit computing power requirement.

[0017] In combination with the first aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, including:

[0018] The control terminal stores the first sub-sample on the edge device, and segments and identifies the first sub-sample by using an image segmentation algorithm;

[0019] The control terminal performs a preprocessing operation on the local device according to the segmentation and recognition results of the edge side device, controls the local device to perform a preliminary analysis on the first subsample according to a preset algorithm, and sends the analysis result to the edge side device;

[0020] After receiving the preliminary analysis results, the edge device further refines the processing, extracts key features, and shares the key features with the cloud device.

[0021] In combination with the first aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample and uploads the second subsample to the cloud device for processing, further comprising:

[0022] After the cloud device completes the processing, it returns the final processing results to the control terminal, which then integrates these results with the processing results of the local device and edge device to form a complete inference result.

[0023] In combination with the first aspect, in some embodiments, the system further includes: a data integration terminal, the control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, including:

[0024] After receiving the preliminary analysis results, the edge device further refines the results, extracts key features, and uploads the key features to the data integration terminal;

[0025] The local device extracts the key features corresponding to the second sub-sample and uploads the key features to the data integration terminal;

[0026] The data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration results to the cloud device.

[0027] In combination with the first aspect, in some implementations, the data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration result to the cloud device, including:

[0028] The data integration terminal performs weighted averaging on key features through a preset algorithm, and trains the deep learning model based on the integration results to optimize the model parameters.

[0029] In conjunction with the first aspect, in some embodiments, the method further comprises:

[0030] The control terminal adjusts the data processing strategy of local devices and edge devices according to the model performance after training of cloud devices, and local devices and edge devices process data with maximum computing power.

[0031] In a second aspect, an embodiment of the present application proposes an AI cluster training system using distributed energy, the system includes a control terminal, an edge device, a local device, and a cloud device, and the method includes:

[0032] The control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling;

[0033] The control terminal allocates target inference samples based on the predicted total computing power requirements and meets the following requirements:

[0034] When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target inference sample so that the target inference sample forms a first sub-sample and a second sub-sample;

[0035] The control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing;

[0036] The control terminal obtains the processing results of the cloud device and integrates the processing results with the processing results of the local device and the edge device.

[0037] In conjunction with the second aspect, in some implementations, the control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on the result of the preliminary sampling, including:

[0038] The control terminal obtains a target reasoning sample, obtains a prediction sample from the target reasoning sample based on a random sampling method, and obtains a corresponding relationship between the prediction sample and the target reasoning sample;

[0039] The control terminal imports the prediction sample into the training device, and obtains the unit computing power requirement corresponding to the prediction sample based on the training result and training time fed back by the training device;

[0040] The control terminal determines the predicted total computing power requirement of the target sample based on the correspondence between the predicted sample and the target inference sample, as well as the unit computing power requirement.

[0041] In conjunction with the second aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, including:

[0042] The control terminal stores the first sub-sample on the edge device, and segments and identifies the first sub-sample by using an image segmentation algorithm;

[0043] The control terminal performs a preprocessing operation on the local device according to the segmentation and recognition results of the edge side device, controls the local device to perform a preliminary analysis on the first subsample according to a preset algorithm, and sends the analysis result to the edge side device;

[0044] After receiving the preliminary analysis results, the edge device further refines the processing, extracts key features, and shares the key features with the cloud device.

[0045] In conjunction with the second aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, further comprising:

[0046] After the cloud device completes the processing, it returns the final processing results to the control terminal, which then integrates these results with the processing results of the local device and edge device to form a complete inference result.

[0047] In combination with the first aspect, in some embodiments, the system further includes: a data integration terminal, the control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, including:

[0048] After receiving the preliminary analysis results, the edge device further refines the results, extracts key features, and uploads the key features to the data integration terminal;

[0049] The local device extracts the key features corresponding to the second sub-sample and uploads the key features to the data integration terminal;

[0050] The data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration results to the cloud device.

[0051] In conjunction with the second aspect, in some implementations, the data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration result to the cloud device, including:

[0052] The data integration terminal performs weighted averaging on key features through a preset algorithm, and trains the deep learning model based on the integration results to optimize the model parameters.

[0053] In conjunction with the second aspect, in some embodiments, the method further comprises:

[0054] The control terminal adjusts the data processing strategy of local devices and edge devices according to the model performance after training of cloud devices, and local devices and edge devices process data with maximum computing power.

[0055] A third aspect of an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0056] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method proposed in the first aspect of the embodiment of the present invention.

[0057] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the embodiment of the present invention.

[0058] In summary, the above method and system have the following technical effects:

[0059] The method proposed in the embodiment of the present application, first, the control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling, then, the control terminal allocates the target reasoning sample based on the predicted total computing power requirement, and satisfies: when the predicted total computing power is greater than the rated computing power of the edge side device and the local device, the control terminal divides the target reasoning sample so that the target reasoning sample forms a first sub-sample and a second sub-sample, then, the control terminal controls the local device and the edge side device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, then, the control terminal obtains the processing result of the cloud device, and integrates the processing result with the processing result of the local device and the edge side device. An AI cluster training method using distributed energy proposed in the embodiment of the present application can effectively utilize distributed energy and improve the efficiency and accuracy of AI cluster training. By reasonably allocating computing tasks between local devices, edge side devices and cloud devices, the computing resources of each device can be fully utilized, data transmission time can be reduced, and latency can be reduced. In addition, by adopting image segmentation algorithms and deep learning model training, the accuracy and speed of data processing can be further improved. Ultimately, by integrating the results processed by different devices, more comprehensive and accurate inference results can be obtained to meet application requirements in complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of an AI cluster training method using distributed energy proposed in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The embodiment of the present application proposes an AI cluster training method using distributed energy, which is applicable to an AI cluster training system using distributed energy. The system includes a control terminal, an edge device, a local device, and a cloud device. Please refer to Figure 1 , the method comprises the following steps:

[0063] S101: The control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling.

[0064] It can be understood that in the present invention, the control terminal is responsible for obtaining the target reasoning samples. First, the control terminal will perform a series of operations to ensure that the target reasoning samples can be successfully obtained. These operations may include communicating with a data storage device, retrieving the required samples, or receiving sample data from an external data source. Once the control terminal successfully obtains the target reasoning samples, the next step is to perform preliminary sampling on these samples.

[0065] Preliminary sampling refers to the process of selecting a portion of representative samples from the target inference sample. This step is crucial because it will directly affect the accuracy and efficiency of subsequent processing. During the preliminary sampling process, the control terminal may use various sampling techniques, such as random sampling, stratified sampling, or cluster sampling. Through these techniques, the control terminal can ensure that the selected samples can reflect the characteristics and distribution of the entire target inference sample as much as possible.

[0066] Specifically, as an implementation manner, step S101 may include the following steps:

[0067] S1011: The control terminal obtains a target reasoning sample, obtains a prediction sample from the target reasoning sample based on a random sampling method, and obtains a corresponding relationship between the prediction sample and the target reasoning sample.

[0068] In this system, the control terminal is responsible for performing a series of operations to ensure the effective processing of the target inference samples. First, the control terminal obtains the target inference samples from the storage device or data source. These samples are predefined data sets for subsequent inference and analysis processes. After obtaining the target inference samples, the control terminal will use a random sampling method to select a portion of them as prediction samples.

[0069] S1012: The control terminal imports the prediction sample into the training device, and obtains the unit computing power requirement corresponding to the prediction sample based on the training result and training time fed back by the training device.

[0070] It can be understood that the control terminal is responsible for importing the prediction samples into the training device for processing. Specifically, the control terminal will transmit the prediction samples to the training device. After receiving these samples, the training device will calculate and analyze them according to the preset training model and algorithm. During the training process, the training device will provide real-time feedback on the training results, which include but are not limited to key indicators such as the accuracy and loss value of the model. At the same time, the control terminal will also monitor the training time of the training device, that is, the time required from the start of training to the completion of training.

[0071] S1013: The control terminal determines the predicted total computing power requirement of the target sample based on the correspondence between the predicted sample and the target inference sample, as well as the unit computing power requirement.

[0072] It is understandable that when the control terminal calculates the predicted total computing power requirement of the target sample, it will comprehensively consider the correspondence between the predicted sample and the target reasoning sample. This correspondence refers to the similarity or correlation between the predicted sample and the target reasoning sample in terms of characteristics, structure or behavior. By analyzing this correspondence, the control terminal can better understand the characteristics of the target reasoning sample and thus more accurately predict its total computing power requirement.

[0073] S102: The control terminal allocates target inference samples based on the predicted total computing power demand and satisfies:

[0074] When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target inference sample so that the target inference sample forms a first sub-sample and a second sub-sample.

[0075] It can be understood that when the total computing power required for the prediction exceeds the rated computing power that the edge-side device and the local device can provide, the control terminal will take measures to split the target reasoning sample. Specifically, the control terminal will split the target reasoning sample into two parts, forming a first sub-sample and a second sub-sample. This segmentation operation is to ensure that the computing requirements of each sub-sample are within the processing capabilities of the edge-side device and the local device, so that the reasoning process can proceed smoothly on these devices. In this way, the control terminal effectively solves the problem of insufficient computing power and ensures the smooth execution of the reasoning task.

[0076] S103: The control terminal controls the local device and the edge side device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing.

[0077] It can be understood that the control terminal is responsible for commanding and managing local devices and edge devices to perform corresponding processing operations on the first sub-sample. During this process, the control terminal will ensure that the local device and the edge device work together to efficiently complete the processing task of the first sub-sample. After the processing is completed, the control terminal will also upload the second sub-sample to the cloud device so that the cloud device can receive the data that needs further processing. This upload process will ensure that the second sub-sample can be received by the cloud device and perform subsequent processing, thereby achieving efficient collaboration of the entire system and smooth data processing.

[0078] Specifically, in this embodiment, the control terminal stores the first sub-sample on the edge side device, and uses the image segmentation algorithm to segment and identify the first sub-sample. Then, the control terminal performs preprocessing operations on the local device according to the segmentation and identification results of the edge side device, and controls the local device to perform preliminary analysis on the first sub-sample according to a preset algorithm, and sends the analysis results to the edge side device. Then, after receiving the preliminary analysis results, the edge side device further refines the processing, extracts key features, and shares the key features with the cloud device.

[0079] Optionally, after the cloud device completes processing, it returns the final processing results to the control terminal, and the control terminal then integrates these results with the processing results of the local device and the edge device to form a complete inference result.

[0080] In some embodiments, the system further includes: a data integration terminal for accelerating data exchange. In this embodiment, after receiving the preliminary analysis results, the edge side device further refines the processing, extracts key features, and uploads the key features to the data integration terminal. Then, the data integration terminal integrates the key features provided by the local device and the edge side device provided by the edge side device, and sends the integration results to the cloud device.

[0081] It is understandable that after receiving the preliminary analysis results, the edge side device will perform further refinement. Specifically, the edge side device will extract key features from the preliminary analysis results, which are important elements that can represent the core information of the data. After the extraction is completed, the edge side device will upload these key features to the data integration terminal. After receiving these key features, the data integration terminal will proceed to the next step. It will not only integrate the key features provided by the local device, but also integrate other relevant data provided by the edge side device. Through this integration, the data integration terminal can fuse data from different sources to obtain more comprehensive and accurate information. After the integration is completed, the data integration terminal will send the integration results to the cloud device. After receiving the integration results, the cloud device can perform further analysis and processing to better utilize these data for decision support or other applications.

[0082] Optionally, the data integration terminal performs weighted averaging processing on key features through a preset algorithm, and trains a deep learning model based on the integration results to optimize model parameters.

[0083] S104: The control terminal obtains the processing results of the cloud device, and integrates the processing results with the processing results of the local device and the edge device.

[0084] In this system, the control terminal is responsible for receiving the processing results from the cloud device. These processing results are obtained by the cloud device after analyzing and calculating the received data. After obtaining these results, the control terminal will integrate them with the processing results of the local device and the edge device. During the integration process, the control terminal will comprehensively consider various factors, such as the real-time nature, accuracy and processing power of the data, to ensure that the final integration result can meet the needs of the system. In this way, the control terminal can make full use of the computing resources of the cloud, local and edge devices to improve the processing efficiency and performance of the entire system.

[0085] Optionally, in some implementations, the control terminal adjusts the data processing strategies of the local device and the edge device according to the model performance after training of the cloud device, and the local device and the edge device process data with maximum computing power.

[0086] The method proposed in the embodiment of the present application, first, the control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling, then, the control terminal allocates the target reasoning sample based on the predicted total computing power requirement, and satisfies: when the predicted total computing power is greater than the rated computing power of the edge side device and the local device, the control terminal divides the target reasoning sample so that the target reasoning sample forms a first sub-sample and a second sub-sample, then, the control terminal controls the local device and the edge side device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, then, the control terminal obtains the processing result of the cloud device, and integrates the processing result with the processing result of the local device and the edge side device. An AI cluster training method using distributed energy proposed in the embodiment of the present application can effectively utilize distributed energy and improve the efficiency and accuracy of AI cluster training. By reasonably allocating computing tasks between local devices, edge side devices and cloud devices, the computing resources of each device can be fully utilized, data transmission time can be reduced, and latency can be reduced. In addition, by adopting image segmentation algorithms and deep learning model training, the accuracy and speed of data processing can be further improved. Ultimately, by integrating the results processed by different devices, more comprehensive and accurate inference results can be obtained to meet application requirements in complex scenarios.

[0087] Based on the same inventive concept, the embodiment of the present application proposes an AI cluster training system using distributed energy, the system includes a control terminal, an edge device, a local device and a cloud device, and the method includes:

[0088] The control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling;

[0089] The control terminal allocates target inference samples based on the predicted total computing power requirements and meets the following requirements:

[0090] When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target inference sample so that the target inference sample forms a first sub-sample and a second sub-sample;

[0091] The control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing;

[0092] The control terminal obtains the processing results of the cloud device and integrates the processing results with the processing results of the local device and the edge device.

[0093] In conjunction with the second aspect, in some implementations, the control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on the result of the preliminary sampling, including:

[0094] The control terminal obtains a target reasoning sample, obtains a prediction sample from the target reasoning sample based on a random sampling method, and obtains a corresponding relationship between the prediction sample and the target reasoning sample;

[0095] The control terminal imports the prediction sample into the training device, and obtains the unit computing power requirement corresponding to the prediction sample based on the training result and training time fed back by the training device;

[0096] The control terminal determines the predicted total computing power requirement of the target sample based on the correspondence between the predicted sample and the target inference sample, as well as the unit computing power requirement.

[0097] In conjunction with the second aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, including:

[0098] The control terminal stores the first sub-sample on the edge device, and segments and identifies the first sub-sample by using an image segmentation algorithm;

[0099] The control terminal performs a preprocessing operation on the local device according to the segmentation and recognition results of the edge side device, controls the local device to perform a preliminary analysis on the first subsample according to a preset algorithm, and sends the analysis result to the edge side device;

[0100] After receiving the preliminary analysis results, the edge device further refines the processing, extracts key features, and shares the key features with the cloud device.

[0101] In conjunction with the second aspect, in some implementations, the control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, further comprising:

[0102] After the cloud device completes the processing, it returns the final processing results to the control terminal, which then integrates these results with the processing results of the local device and edge device to form a complete inference result.

[0103] In combination with the first aspect, in some embodiments, the system further includes: a data integration terminal, the control terminal controls the local device and the edge device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, including:

[0104] After receiving the preliminary analysis results, the edge device further refines the results, extracts key features, and uploads the key features to the data integration terminal;

[0105] The local device extracts the key features corresponding to the second sub-sample and uploads the key features to the data integration terminal;

[0106] The data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration results to the cloud device.

[0107] In conjunction with the second aspect, in some implementations, the data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration result to the cloud device, including:

[0108] The data integration terminal performs weighted averaging on key features through a preset algorithm, and trains the deep learning model based on the integration results to optimize the model parameters.

[0109] In conjunction with the second aspect, in some embodiments, the method further comprises:

[0110] The control terminal adjusts the data processing strategy of local devices and edge devices according to the model performance after training of cloud devices, and local devices and edge devices process data with maximum computing power.

[0111] A third aspect of an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0112] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method proposed in the first aspect of the embodiment of the present invention.

[0113] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in the first aspect of the embodiment of the present invention.

[0114] The system proposed in the embodiment of the present application, first, the control terminal obtains the target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines the predicted total computing power requirement of the target sample based on the result of the preliminary sampling, then, the control terminal allocates the target reasoning sample based on the predicted total computing power requirement, and satisfies: when the predicted total computing power is greater than the rated computing power of the edge side device and the local device, the control terminal divides the target reasoning sample so that the target reasoning sample forms a first sub-sample and a second sub-sample, then, the control terminal controls the local device and the edge side device to process the first sub-sample, and uploads the second sub-sample to the cloud device for processing, then, the control terminal obtains the processing result of the cloud device, and integrates the processing result with the processing result of the local device and the edge side device. An AI cluster training method using distributed energy proposed in the embodiment of the present application can effectively utilize distributed energy and improve the efficiency and accuracy of AI cluster training. By reasonably allocating computing tasks between local devices, edge side devices and cloud devices, the computing resources of each device can be fully utilized, data transmission time can be reduced, and latency can be reduced. In addition, by adopting image segmentation algorithms and deep learning model training, the accuracy and speed of data processing can be further improved. Ultimately, by integrating the results processed by different devices, more comprehensive and accurate inference results can be obtained to meet application requirements in complex scenarios.

[0115] Based on the same inventive concept, an embodiment of the present application further proposes an electronic device, the electronic device comprising:

[0116] At least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the automatic overheating protection method for a universal testing machine based on an embodiment of the present application.

[0117] In addition, to achieve the above-mentioned purpose, an embodiment of the present application further proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the automatic overheating protection method based on a universal testing machine of an embodiment of the present application.

[0118] The following is a detailed introduction to the various components of electronic equipment:

[0119] The processor is the control center of the electronic device, which can be a processor or a general term for multiple processing elements. For example, the processor is one or more central processing units (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSP), or one or more field programmable gate arrays (field programmable gate arrays, FPGA).

[0120] Optionally, the processor can perform various functions of the electronic device by running or executing a software program stored in the memory, and calling data stored in the memory.

[0121] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0122] Optionally, the memory may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be integrated with the processor, or may exist independently and be coupled to the processor through an interface circuit of the electronic device, and the embodiments of the present invention do not specifically limit this.

[0123] A transceiver is used to communicate with a network device or a terminal device.

[0124] Optionally, the transceiver may include a receiver and a transmitter, wherein the receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0125] Optionally, the transceiver may be integrated with the processor, or may exist independently and be coupled to the processor via an interface circuit of the router, which is not specifically limited in the embodiment of the present invention.

[0126] In addition, the technical effects of the electronic device can refer to the technical effects of the data transmission method in the above method embodiment, which will not be repeated here.

[0127] It should be understood that the processor in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0128] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0129] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When a computer instruction or computer program is loaded or executed on a computer, a process or function according to an embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0130] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0131] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0132] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0133] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

Claims

1. An AI cluster training method using distributed energy, characterized in that: Applicable to an AI cluster training system using distributed energy, the system includes a control terminal, an edge device, a local device and a cloud device, and the method includes: The control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on a result of the preliminary sampling; The control terminal allocates the target reasoning samples based on the predicted total computing power requirement and satisfies: When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target reasoning sample so that the target reasoning sample forms a first sub-sample and a second sub-sample; The control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing; The control terminal obtains the processing result of the cloud device, and integrates the processing result with the processing results of the local device and the edge side device.

2. The AI ​​cluster training method using distributed energy according to claim 1, characterized in that: The control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on a result of the preliminary sampling, including: The control terminal obtains the target reasoning sample, obtains a prediction sample from the target reasoning sample based on a random sampling method, and obtains a corresponding relationship between the prediction sample and the target reasoning sample; The control terminal imports the predicted sample into the training device, and obtains the unit computing power requirement corresponding to the predicted sample based on the training result and training time fed back by the training device; The control terminal determines the predicted total computing power requirement of the target sample based on the corresponding relationship between the predicted sample and the target inference sample, and the unit computing power requirement.

3. The AI ​​cluster training method using distributed energy according to claim 2, characterized in that: The control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, including: The control terminal stores the first sub-sample on the edge side device, and segments and identifies the first sub-sample using an image segmentation algorithm; The control terminal performs a preprocessing operation on the local device according to the segmentation and identification results of the edge side device, controls the local device to perform a preliminary analysis on the first subsample according to a preset algorithm, and sends the analysis result to the edge side device; After receiving the preliminary analysis results, the edge device further refines the processing, extracts key features, and shares the key features with the cloud device.

4. The AI ​​cluster training method using distributed energy according to claim 3 is characterized in that: The control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing, further comprising: After the cloud device completes the processing, it returns the final processing results to the control terminal, and the control terminal then integrates these results with the processing results of the local device and the edge device to form a complete inference result.

5. The AI ​​cluster training method using distributed energy according to claim 3 is characterized in that: The system further includes: a data integration terminal, wherein the control terminal controls the local device and the edge device to process the first subsample and uploads the second subsample to the cloud device for processing, including: After receiving the preliminary analysis results, the edge side device further refines the results, extracts key features, and uploads the key features to the data integration terminal; The local device extracts the key feature corresponding to the second sub-sample, and uploads the key feature to the data integration terminal; The data integration terminal integrates the key features provided by the local device and the edge-side features provided by the edge-side device, and sends the integration result to the cloud device.

6. The AI ​​cluster training method using distributed energy according to claim 5, characterized in that: The data integration terminal integrates the key features provided by the local device and the edge device provided by the edge device, and sends the integration result to the cloud device, including: The data integration terminal performs weighted averaging processing on the key features through a preset algorithm, and trains a deep learning model based on the integration results to optimize model parameters.

7. The AI ​​cluster training method using distributed energy according to claim 5, characterized in that: The method further comprises: The control terminal adjusts the data processing strategies of the local device and the edge device according to the model performance after training of the cloud device, and the local device and the edge device perform data processing with maximum computing power.

8. An AI cluster training system using distributed energy, characterized in that: The system includes a control terminal, an edge device, a local device, and a cloud device, and the system is configured as follows: The control terminal obtains a target reasoning sample, performs preliminary sampling on the target reasoning sample, and determines a predicted total computing power requirement of the target sample based on a result of the preliminary sampling; The control terminal allocates the target reasoning samples based on the predicted total computing power requirement and satisfies: When the predicted total computing power is greater than the rated computing power of the edge device and the local device, the control terminal splits the target reasoning sample so that the target reasoning sample forms a first sub-sample and a second sub-sample; The control terminal controls the local device and the edge device to process the first subsample, and uploads the second subsample to the cloud device for processing; The control terminal obtains the processing result of the cloud device, and integrates the processing result with the processing results of the local device and the edge side device.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; And, a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute the method as set forth in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the method as claimed in any one of claims 1 to 7 is implemented.

Images