Model algorithm operation method, device and storage medium

Abstract

The present application relates to the technical field of machine learning, and discloses a model algorithm operation method, a device and a storage medium. The method of the present application comprises: acquiring an algorithm task to be executed; allocating a corresponding target computing power node to said algorithm task; and controlling a demand node corresponding to said algorithm task to send a task execution parameter to the target computing power node, so that the target computing power node executes said algorithm task on the basis of the task execution parameter.

Description

Model algorithm operation method, device and storage medium

[0001] Cross-references

[0002] This application claims priority to the Chinese patent application filed with the China Patent Office on November 15, 2023, with application number 202311530436.7 and invention name “Model Algorithm Operation Method, Device and Storage Medium”. The entire contents of the application are incorporated by reference into this application. Technical Field

[0003] The present application relates to the field of machine learning technology, and in particular to a model algorithm operation method, device and storage medium. Background Art

[0004] Nowadays, the model algorithms of artificial intelligence (AI) applications are mostly calculated in the cloud. However, AI applications have high computing power requirements and extremely high performance requirements on the cloud, which puts a lot of pressure on the cloud. Performing machine learning on the edge and end side (i.e., running the corresponding model algorithms of machine learning) can improve the reliability and response speed of local devices, reduce the data bandwidth uploaded to the cloud, and ensure the security of local data. Therefore, AI computing has gradually developed towards the edge and end side.

[0005] Deploying machine learning for various components and devices at the edge requires close business collaboration between edge and end-side devices. However, end-side applications suffer from scenario fragmentation. For example, the power supply system in a communication site computer room is typically a heterogeneous system consisting of power supplies, batteries, temperature control, and dynamic environment equipment from different manufacturers. The scenarios are complex, and there are many versions of various devices and diverse interface protocols. This leads to extremely complex interactions between the edge and end-side, making it very difficult to meet user personalized needs, independently implement new services, and upgrade and maintain them, thereby limiting the development of end-side machine learning.

[0006] Summary of the Invention

[0007] The main purpose of this application is to provide a model algorithm operation method, device and storage medium.

[0008] To achieve the above-mentioned purpose, an embodiment of the present application provides a model algorithm operation method, which is applied to a power supply system, wherein the power supply system is provided with interconnected demand nodes and computing power nodes, the demand node is an intelligent node device in the power supply system that can execute the model algorithm, and the computing power node is a demand node in an idle state; the model algorithm operation method includes: obtaining the algorithm task to be executed; allocating the corresponding target computing power node to the algorithm task to be executed; controlling the demand node corresponding to the algorithm task to be executed to send the task execution parameters to the target computing power node, so that the target computing power node executes the algorithm task to be executed according to the task execution parameters.

[0009] In addition, to achieve the above-mentioned purpose, an embodiment of the present application also proposes a model algorithm running device, which is applied to a power supply system, wherein the power supply system is provided with interconnected demand nodes and computing power nodes, the demand node is an intelligent node device in the power supply system that can execute the model algorithm, and the computing power node is a demand node in an idle state; the model algorithm running device includes: an acquisition module for acquiring the algorithm task to be executed; an allocation module for allocating the corresponding target computing power node to the algorithm task to be executed; and a control module for controlling the demand node corresponding to the algorithm task to be executed to send the task execution parameters to the target computing power node, so that the target computing power node executes the algorithm task to be executed according to the task execution parameters.

[0010] In addition, to achieve the above-mentioned purpose, an embodiment of the present application also proposes a model algorithm running device, which includes: a processor, a memory, and a model algorithm running program stored on the memory and capable of running on the processor. When the model algorithm running program is executed by the processor, the steps of the model algorithm running method described above are implemented.

[0011] In addition, to achieve the above-mentioned purpose, an embodiment of the present application also proposes a computer-readable storage medium, on which a model algorithm running program is stored. When the model algorithm running program is executed, the steps of the model algorithm running method described above are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIG1 is a schematic diagram of the structure of an electronic device in a hardware operating environment according to an embodiment of the present application;

[0013] FIG2 is a flow chart of the first embodiment of the model algorithm operation method of the present application;

[0014] FIG3 is a flow chart of a second embodiment of the model algorithm operation method of the present application;

[0015] FIG4 is a flow chart of a third embodiment of the model algorithm operation method of the present application;

[0016] FIG5 is a flow chart of a fourth embodiment of the model algorithm operation method of the present application;

[0017] FIG6 is a structural block diagram of the first embodiment of the model algorithm operation device of the present application. DETAILED DESCRIPTION

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

[0019] Refer to Figure 1, which is a schematic diagram of the model algorithm running device structure of the hardware running environment involved in the embodiment of the present application.

[0020] As shown in Figure 1, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and optionally the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0021] Those skilled in the art will appreciate that the structure shown in FIG1 does not limit the electronic device and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0022] As shown in FIG1 , the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a model algorithm running program.

[0023] In the electronic device shown in Figure 1, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of this application can be set in the model algorithm running device, and the electronic device calls the model algorithm running program stored in the memory 1005 through the processor 1001, and executes the model algorithm running method provided in the embodiment of this application.

[0024] The present application provides a model algorithm operation method, with reference to Figure 2, which is a flow chart of a first embodiment of a model algorithm operation method of the present application. In this embodiment, the model algorithm operation method includes the following steps.

[0025] Step S10: Obtain the algorithm task to be executed.

[0026] It should be noted that the executing entity of this embodiment can be the power supply system or the model algorithm running device deployed in the power supply system. The model algorithm running device can be a device that can monitor and schedule each device in the power supply system, such as a centralized controller (Concentrated Supervision Unit, CSU). Of course, it can also be other devices that can achieve the same or similar functions. This embodiment does not limit this. In this embodiment and the following embodiments, the model algorithm running method of this application is explained by taking the model algorithm running device as an example.

[0027] It should be noted that the power supply system mentioned in this embodiment can be an intelligent power supply system in related fields such as communication power supply, energy storage system, photovoltaic power generation, etc., such as an intelligent power supply system that supplies power to communication base stations. The power supply system is provided with interconnected demand nodes and computing power nodes. The demand nodes are intelligent node devices in the power supply system that can execute model algorithms, and the computing power nodes are demand nodes in an idle state.

[0028] Among them, the intelligent node device can be a device equipped with a computing power core (such as a central processing unit (CPU), a microcontroller unit (MCU) and other core processors) in the power supply system, such as a rectifier, an intelligent power supply, etc.; the intelligent node device that can execute the model algorithm can be an intelligent node device that has been modified so that it can run the model algorithm, for example: the model algorithm code (C language code or other types of code) corresponding to the model algorithm is burned into the computing power core of the intelligent node device.

[0029] In order to avoid excessive modifications to the power supply system, the connection between the demand nodes, computing power nodes and model algorithm running devices can utilize the connection method of the power supply system itself. For example, if the devices in the power supply system use a bus (such as the Controller Area Network (CAN) bus) for connection and communication, then the demand nodes, computing power nodes and model algorithm running devices can also use the bus for communication.

[0030] In actual use, in order not to affect the power supply system and to ensure reliability, a redundant design is generally adopted. That is, on the basis of ensuring the normal operation of the power supply system, an additional part of intelligent node devices are added. In order to avoid the intelligent node devices running at low load and low utilization, the model algorithm running device can control the intelligent node devices to rotate and perform power supply tasks. At this time, some intelligent node devices will inevitably be idle.

[0031] For example, the power supply system (CSU) can control some node devices, such as rectifiers and batteries, to rotate between working, standby, or hibernation according to business needs, thereby reducing the number of working node devices to achieve optimal system energy efficiency or extend the service life of node devices. Specific control can be as follows.

[0032] (1) Rectifier: The CSU dynamically controls the rectifiers to rotate their operation. The working rectifiers run at the most efficient load point, improving energy conversion efficiency; the non-working rectifiers rest, extending the device life.

[0033] (2) Smart Battery: The CSU dynamically controls the batteries to rotate their work. The working battery modules operate at the highest efficiency point (if the battery has a built-in DC-DC converter (DC / DC), the overall energy efficiency of the battery cell charge and discharge efficiency and DC / DC conversion efficiency must be controlled to maximize the efficiency). The inactive batteries rest, extending the device life.

[0034] (3) Other devices: such as DC / DC, direct current to alternating current converter (DC / AC), etc., also perform sleep or rotation control.

[0035] During this process, the idle resources such as computing power, memory, and network of the non-working node devices (only their power conversion parts are not working) are used to run the model algorithm.

[0036] On this basis, in order to avoid affecting the normal operation of the power supply system, it is necessary to avoid complex calculations of intelligent node devices that are performing power supply tasks. Therefore, the demand nodes can be further classified, and some of the demand nodes that are in idle state can be used as computing power nodes. At this time, when the demand nodes need to perform complex calculations, they can not execute them directly, but submit the algorithm tasks to the model algorithm running device, and the model algorithm running device will assign them the corresponding computing power nodes for execution.

[0037] Of course, the computing power of the computing node itself may not be completely sufficient. In order to arrange the calculation reasonably, the computing power node can also generate corresponding algorithm tasks when it needs to perform complex calculations, and submit them to the model algorithm running device, which will then distribute them uniformly.

[0038] During actual processing, the model algorithm running device may connect to multiple demand nodes at the same time. At this time, the model algorithm running device can process the algorithm tasks submitted by each demand node in turn. The algorithm tasks to be executed can be the algorithm tasks that the model algorithm running device is currently processing (i.e., allocating computing power nodes).

[0039] Step S20: Allocate a corresponding target computing power node to the algorithm task to be executed.

[0040] It can be understood that after the model algorithm running device obtains the algorithm task to be executed, it needs to allocate the corresponding computing power node for processing. At this time, the model algorithm running device can scan the operating status of each demand node, determine which ones are currently serving as computing power nodes, and then select the computing power node that processes the algorithm task to be executed from the computing power nodes as the target computing power node.

[0041] Among them, in order to avoid waiting too long for the execution of algorithm tasks, when allocating the corresponding target computing power node for the algorithm task to be executed, the computing power node that is not currently executing other algorithm tasks can be allocated as the target computing power node.

[0042] Step S30: Control the demand node corresponding to the algorithm task to be executed to send the task execution parameters to the target computing power node, so that the target computing power node executes the algorithm task to be executed according to the task execution parameters.

[0043] It can be understood that after the corresponding target computing power node is assigned to the algorithm task to be executed, the demand node corresponding to the algorithm task to be executed can be notified, so that the demand node corresponding to the algorithm task to be executed will send the task execution parameters required to execute the algorithm task to be executed to the target computing power node, and the target computing power node will execute the algorithm task to be executed according to the task execution parameters, and after the execution is completed, it will feedback the results generated after the execution of the algorithm task to be executed to the demand node corresponding to the algorithm task to be executed.

[0044] The demand node corresponding to the algorithm task to be executed may be the demand node for submitting the algorithm task to be executed to the model algorithm running device. The task execution parameters may be various parameters involved in the calculation of the algorithm task to be executed.

[0045] This embodiment obtains pending algorithm tasks, assigns corresponding target computing nodes to them, and controls the demand nodes corresponding to the pending algorithm tasks to send task execution parameters to the target computing nodes, so that the target computing nodes execute the pending algorithm tasks according to the task execution parameters. Because the power system is modified and idle computing nodes are used to execute the model algorithm, and this process does not involve frequent edge-side and end-side interactions, the model algorithm can be set directly for execution on the end-side while also being very easy to maintain.

[0046] Refer to FIG3 , which is a flow chart of a second embodiment of a model algorithm operation method of the present application.

[0047] Based on the above-mentioned first embodiment, the step S10 of the model algorithm running method of this embodiment includes: Step S101: Detect whether there is an algorithm task in the task list.

[0048] It should be noted that the task list can be a list for storing algorithm tasks to be executed. In order to ensure that the submitted algorithm tasks can be executed sequentially, the task list can be constructed based on an ordered data structure (such as a queue, a doubly linked list, etc.) to ensure that the algorithm tasks in the task list can be processed sequentially based on the first-in-first-out principle.

[0049] In actual use, since the model algorithm running device may connect to multiple demand nodes at the same time, that is, multiple demand nodes submit algorithm tasks to the model algorithm running device, in order to avoid data loss, the model algorithm running device can store the algorithm tasks in the order of submission when receiving the algorithm tasks submitted by the demand nodes, and then process them in the sort order of the task list.

[0050] Step S102: If it exists, the algorithm task ranked first in the task list is used as the algorithm task to be executed.

[0051] It should be noted that the sorting order of the algorithm tasks in the task list is the order in which the algorithm tasks are processed.

[0052] If there is an algorithm task in the task list, it means that a demand node has submitted the algorithm task. The model algorithm running device needs to allocate the corresponding computing power node to the algorithm task for processing to ensure that the algorithm task can be executed normally. The algorithm task ranked first in the task list can be used as the algorithm task to be executed.

[0053] This embodiment detects whether an algorithm task exists in a task list, which is used to store pending algorithm tasks. If so, the algorithm task ranked first in the task list is selected as the pending algorithm task. Upon receiving algorithm tasks submitted by a demand node, the algorithm tasks are sequentially stored in the task list and subsequently executed according to the sort order in the task list. This ensures that the algorithm tasks submitted by each demand node can be executed in the correct order.

[0054] Refer to FIG4 , which is a flow chart of a third embodiment of a model algorithm operation method of the present application.

[0055] Based on the above second embodiment, the step S10 of the model algorithm running method of this embodiment includes: step S101 ′: detecting whether there is an algorithm task in the priority task list.

[0056] It should be noted that although in principle the algorithm tasks submitted by each demand node need to be executed in the order of submission, in actual processing, it is inevitable that some algorithm tasks will be more important and need to be executed first. In order to ensure that such algorithm tasks can be executed first, the set task list can include a priority task list and a secondary task list. The priority task list stores algorithm tasks with high execution priority, and the secondary task list stores algorithm tasks with low execution priority.

[0057] Among them, algorithm tasks with high execution priority can be more important and need to be executed first; algorithm tasks with low execution priority can be algorithm tasks that do not need to be executed first but are executed sequentially. Both the priority task list and the secondary task list can be constructed based on an ordered data structure. The order in which each algorithm task is executed is the order in which each algorithm task is sorted in the priority task list and the secondary task list.

[0058] It is understandable that after setting the priority task list and the secondary task list, in order to ensure that the algorithm tasks with high execution priority are executed first, the order of the detection task list also needs to be adjusted accordingly. At this time, you can first check whether there is an algorithm task in the priority task list.

[0059] In a specific implementation, in order to ensure that the execution priority of each algorithm task can be correctly distinguished, before step S101' described in this embodiment, it may also include: after receiving the algorithm task submitted by the demand node, obtaining the delay requirement value corresponding to the algorithm task; when the delay requirement value is greater than the preset delay threshold, adding the algorithm task to the secondary task list; when the delay requirement value is less than or equal to the preset delay threshold, adding the algorithm task to the priority task list.

[0060] It should be noted that the delay requirement value corresponding to the algorithm task can be used to represent the maximum length of time the algorithm task can wait for delayed execution. For example, if the delay requirement value corresponding to the algorithm task is 30 minutes, it means that the algorithm task needs to be executed after waiting for a maximum of 30 minutes.

[0061] In actual use, the delay requirement can be used to measure the urgency of the algorithm task. The smaller the delay requirement value, the higher the urgency of the algorithm task. The larger the delay requirement value, the lower the urgency of the algorithm task. Therefore, the execution priority of the algorithm task can be determined based on the delay requirement value.

[0062] It is understood that if the delay requirement value is less than or equal to the preset delay threshold, it means that the algorithm task is more urgent and needs to be executed first. Therefore, the algorithm task can be added to the priority task list. If the delay requirement value is greater than the preset delay threshold, it means that the algorithm task is less urgent and does not need to be executed first. Therefore, the algorithm task can be added to the secondary task list.

[0063] When adding an algorithm task to the priority task list or the secondary task list, the algorithm task may be added to the end of the task list, that is, the algorithm task is sorted last in the task list.

[0064] Step S102 ′: If the priority task list is a blank list, then check whether there is an algorithm task in the secondary task list.

[0065] It should be noted that if the priority task list is blank, it means that there is no algorithm task that needs to be processed first. Therefore, it is possible to further check whether there is an algorithm task in the secondary task list. If there is an algorithm task in the priority task list, it means that there is an algorithm task that needs to be processed first. The algorithm task ranked first in the priority task list can be used as the algorithm task to be executed.

[0066] Step S103 ′: if there is an algorithm task in the secondary task list, the algorithm task ranked first in the secondary task list is used as the algorithm task to be executed.

[0067] It can be understood that if the priority task list is a blank list, but there is an algorithm task in the secondary task list, it means that there is no algorithm task with a high execution priority at this time. At this time, each algorithm task can be processed only according to the sorting order of each task in the secondary task list. Therefore, the algorithm task ranked first in the secondary task list can be used as the algorithm task to be executed.

[0068] In a specific implementation, in order to avoid a single algorithm task from waiting for too long, the status of each algorithm task in the secondary task list can be monitored. At this time, before step S101' described in this embodiment, it can also include: detecting whether there is a waiting timeout task in the secondary task list; if there is a waiting timeout task, adding the waiting timeout task to the priority task list.

[0069] It should be noted that the waiting timeout task is an algorithm task, etc., in which the difference between the task creation time and the current time is greater than the delay requirement value.

[0070] Among them, the task creation time can be the time when the algorithm task is created by the demand node, the time when it is submitted to the model algorithm running device, or the time when it is added to the task list (added to the priority task list or the secondary task list).

[0071] For example: if the current time is A, the task creation time corresponding to the algorithm task is B, and the delay requirement value is C, then if AB≥C, it can be determined that the algorithm task is a waiting timeout task.

[0072] In actual use, if there is a waiting timeout task in the secondary task list, it means that some algorithm tasks added to the secondary task list have been waiting for too long and have not been executed. At this time, the algorithm task needs to be processed first. Therefore, its execution priority can be increased and the waiting timeout task can be added to the priority task list.

[0073] Similarly, in order to avoid a single algorithm task waiting time being too long, the status of each algorithm task in the priority task list can also be monitored. At this time, before step S101' described in this embodiment, it can also include: detecting whether there is a waiting timeout task in the priority task list; if there is a waiting timeout task, then increasing the sorting priority of the waiting timeout task in the priority task list.

[0074] It should be noted that if there is a waiting timeout task in the priority task list, it means that some algorithm tasks added to the priority task list have been waiting for too long and have not been executed. At this time, the algorithm task needs to be processed first, and the sorting priority of the waiting timeout task in the priority task list can be increased.

[0075] In actual use, increasing the priority of the waiting timeout task in the priority task list may be to increase the order of the waiting timeout task in the priority task list. The degree of increase in the order may be pre-set by the administrator of the model algorithm running device.

[0076] For example: if the sorting order of the waiting timeout task in the priority task list is increased by one (such as from fifth to fourth); or the sorting order of the waiting timeout task in the priority task list is directly increased to first (such as from fifth to first).

[0077] In the specific implementation, although the task list constructed by data structures such as queues is more convenient to access, due to its underlying implementation (the underlying implementation is an array or collection), if the sorting order of the elements needs to be changed, the entire structure needs to be modified, which is more complicated and time-consuming. For the doubly linked list data structure, if the sorting order of the elements is modified, only the pointers in some of the elements need to be modified, and the modification speed is faster. Therefore, in order to facilitate the improvement of the sorting order of algorithm tasks in the task list, a doubly linked list can be used to construct the task list.

[0078] This embodiment detects whether an algorithm task exists in the priority task list. If the priority task list is blank, it then detects whether an algorithm task exists in the secondary task list. If an algorithm task exists in the secondary task list, the algorithm task ranked first in the secondary task list is selected as the algorithm task to be executed. Because both the priority task list and the secondary task list are set, while ensuring that the algorithm tasks submitted by each demand node can be executed sequentially, some high-priority tasks are also allowed to be executed first, which better meets the computing needs of actual scenarios.

[0079] Refer to FIG5 , which is a flow chart of a fourth embodiment of a model algorithm operation method of the present application.

[0080] Based on the above-mentioned first embodiment, the step S20 of the model algorithm running method of this embodiment includes: step S201: obtaining the task execution resource requirements and requirement node types corresponding to the algorithm task to be executed.

[0081] It should be noted that a power supply system may contain a variety of intelligent node devices of different device types. For example, a power supply system may simultaneously contain intelligent node devices of various device types, such as rectifiers, inverters, photovoltaic modules, and intelligent batteries. However, since the computing power cores in intelligent node devices generally have relatively weak computing power and memory resources and cannot undergo extensive modification, different types of intelligent node devices may require different modification methods, and the algorithms that can be executed after modification may also differ. Therefore, when assigning corresponding computing power nodes to demand nodes, it is necessary to assign nodes of the same type as the demand nodes corresponding to the algorithm task to be executed, and to have computing power nodes capable of completing the algorithm task. At this time, in order to correctly assign computing power nodes, the task execution resource requirements and demand node types corresponding to the algorithm task to be executed can be obtained.

[0082] Among them, the task execution resource requirements corresponding to the algorithm task to be executed can be the computing resources required to execute the algorithm task to be executed (such as the required CPU computing power, memory size, etc.), and the demand node type corresponding to the algorithm task to be executed can be the device type of the demand node that submits the algorithm task to be executed.

[0083] For example, multiple AI algorithm modules are embedded (or deployed) in the rectifier device software, such as the status monitoring module a1 that detects whether the fan is blocked, and the algorithm module a2 that predicts the aging degree of a key component. Multiple AI algorithm modules are also embedded in the intelligent battery management system (BMS) software, such as the state of charge (SOC) prediction algorithm module b1, the battery short circuit identification algorithm module b2, and the solar power supply scenario perception module b3. Therefore, if a rectifier proposes an AI algorithm task to be executed at this time, the AI ​​algorithm task needs to be assigned to other rectifiers for execution.

[0084] Step S202: Detect whether there is a computing power node that meets the task execution resource requirements, whose node type is consistent with the required node type, and is in a waiting-to-be-allocated state.

[0085] Step S203: If so, assign the corresponding target computing power node to the algorithm task to be executed.

[0086] It should be noted that in order to facilitate allocation, the model algorithm running device can record the status of each computing power node to distinguish whether the corresponding algorithm task has been assigned to it. At this time, the status of the computing power node that has been assigned the algorithm task can be marked as the allocated status, and the status of the computing power node that has not been assigned the algorithm task or the assigned algorithm task has been completed can be marked as the pending allocation status. Then, when it is necessary to assign the pending algorithm task, the model algorithm running device can detect whether there is a node that meets the task execution resource requirements, the node type is consistent with the required node type, and is in the pending allocation status.

[0087] Among them, the node type of the computing power node can be the device type of the computing power node; if the computing resources of the computing power core in the computing power node are greater than or equal to the task execution resource requirements, it can be determined that the task execution resource requirements are met.

[0088] It can be understood that if there is a computing power node that meets the task execution resource requirements, the node type is consistent with the required node type, and is in a waiting-to-be-allocated state, then the computing power node can be used as the target computing power node corresponding to the algorithm task to be executed.

[0089] In actual use, after allocating the corresponding target computing power node to the algorithm task to be executed, the model algorithm running device can set the state of the target computing power node to the allocated state and notify the demand node corresponding to the algorithm task to be executed. The demand node then sends the identifier (ID) of the algorithm task to be executed and various input parameters (or characteristic parameters) to the target computing power node through the CAN bus network, and the target computing power node executes the algorithm task to be executed. After the execution is completed, the target computing power node feeds back the output result to the demand node corresponding to the algorithm task to be executed through the CAN bus network. After receiving the output result, the demand node corresponding to the algorithm task to be executed will notify the model algorithm running device, and the model algorithm operation will set the state of the target computing power node to the to-be-allocated state.

[0090] If it does not exist, the algorithm task to be executed can be marked as a waiting state, and the process returns to step S202 to continue monitoring.

[0091] In a specific implementation, in order to ensure that the power supply system has sufficient computing power to execute the model algorithm, after step S202 described in this embodiment, it may also include: when the task execution resource requirements are met and the computing power nodes whose node types are consistent with the required node types are in an allocated state, the shared computing power node is set as the target computing power node corresponding to the algorithm task to be executed.

[0092] It should be noted that shared computing nodes can be computing devices that managers of model algorithm running devices can also add to the power system. Shared computing nodes can also be computing nodes that can execute the same model algorithm as demand nodes. Shared computing nodes can also communicate with demand nodes, computing nodes, and model algorithm running devices through the connection method of the power system itself. Different shared computing nodes can be set for different node types, or a unified computing node can be set.

[0093] For example: after burning the codes of all the algorithm modules of the rectifier (fan status monitoring module a1, device aging prediction module a2, etc.) on the shared computing power node, it can be used as the public computing power node of the rectifier; if the codes of all the algorithm modules of the battery (SOC prediction module b1, internal short circuit identification module b2, power supply scenario perception module b3, etc.) are loaded, it can be used as the public computing power node of the rectifier and battery.

[0094] It can be understood that when the task execution resource requirements are met and the computing power nodes whose node types are consistent with the required node types are in an allocated state, it means that there is no computing power node that can execute the algorithm task to be executed at the current moment. At this time, the algorithm task to be executed can be executed by the public computing power node. Therefore, the shared computing power node can be set as the target computing power node corresponding to the algorithm task to be executed.

[0095] Among them, if the computing power and memory resources of the shared computing power node are strong enough, the shared computing power node can be used as an edge computing gateway. In this case, there is no need to set up computing power nodes, and the shared computing power node can directly execute the algorithm tasks submitted by all demand nodes.

[0096] On this basis, shared computing power nodes can be configured in the following three ways.

[0097] (1) No configuration: The idle end devices (i.e., idle demand nodes) in the power system have sufficient computing power, so there is no need to configure shared computing power nodes.

[0098] (2) Low configuration: The computing power of the idle end device is insufficient. In this case, while the idle end device is used as a computing power node, an additional public computing module can be set up.

[0099] (3) High configuration: The idle end device has insufficient computing power, and the shared computing power node has higher computing power and larger memory resources. In this case, the idle end device can no longer be used as a computing power node.

[0100] In specific implementation, the peak demand of the model algorithm used can be compared with the computing power of the idle end device to determine whether the computing power of the idle end device is sufficient.

[0101] This embodiment obtains the task execution resource requirements and required node type corresponding to the pending algorithm task, detects whether there is a computing power node that meets the task execution resource requirements, has the same node type as the required node, and is in a pending allocation state. If so, the corresponding target computing power node is allocated to the pending algorithm task. Because the target computing power node is allocated based on the task execution resource requirements and required node type corresponding to the pending algorithm task, it can be guaranteed that the ultimately allocated target computing power node can normally execute the pending algorithm task.

[0102] In addition, an embodiment of the present application also proposes a storage medium, on which a model algorithm running program is stored. When the model algorithm running program is executed by a processor, the steps of the model algorithm running method described above are implemented.

[0103] Refer to Figure 6, which is a structural block diagram of the first embodiment of the model algorithm running device of the present application.

[0104] As shown in Figure 6, the model algorithm running device proposed in the embodiment of the present application includes: an acquisition module 10, used to obtain the algorithm task to be executed; an allocation module 20, used to allocate the corresponding target computing power node for the algorithm task to be executed; and a control module 30, used to control the demand node corresponding to the algorithm task to be executed to send the task execution parameters to the target computing power node, so that the target computing power node executes the algorithm task to be executed according to the task execution parameters.

[0105] This embodiment obtains pending algorithm tasks, assigns corresponding target computing nodes to them, and controls the demand nodes corresponding to the pending algorithm tasks to send task execution parameters to the target computing nodes, so that the target computing nodes execute the pending algorithm tasks according to the task execution parameters. Because the power system is modified and idle computing nodes are used to execute the model algorithm, and this process does not involve frequent edge-side and end-side interactions, the model algorithm can be set directly for execution on the end-side while also being very easy to maintain.

[0106] In a possible implementation of the present application, the acquisition module 10 is also used to detect whether there is an algorithm task in the task list, and the task list is used to store algorithm tasks to be executed; if it exists, the algorithm task ranked first in the task list is used as the algorithm task to be executed.

[0107] In a possible implementation of the present application, the task list includes a priority task list and a secondary task list, the priority task list is used to store algorithm tasks with high execution priority, and the secondary task list is used to store algorithm tasks with low execution priority; the acquisition module 10 is also used to detect whether there is an algorithm task in the priority task list; if the priority task list is a blank list, it is detected whether there is an algorithm task in the secondary task list; if there is an algorithm task in the secondary task list, the algorithm task ranked first in the secondary task list is used as the algorithm task to be executed.

[0108] In a possible implementation of the present application, the acquisition module 10 is also used to obtain the delay requirement value corresponding to the algorithm task after receiving the algorithm task submitted by the demand node; when the delay requirement value is greater than the preset delay threshold, the algorithm task is added to the secondary task list; when the delay requirement value is less than or equal to the preset delay threshold, the algorithm task is added to the priority task list.

[0109] In a possible implementation of the present application, the acquisition module 10 is also used to detect whether there is a waiting timeout task in the secondary task list, and the waiting timeout task is an algorithm task whose difference between the task creation time and the current time is greater than the delay requirement value; if there is a waiting timeout task, the waiting timeout task is added to the priority task list.

[0110] In a possible implementation of the present application, the acquisition module 10 is further configured to detect whether there is a waiting timeout task in the priority task list; if there is a waiting timeout task, the sorting priority of the waiting timeout task in the priority task list is increased.

[0111] In a possible implementation of the present application, the demand node includes demand nodes of multiple different node types; the allocation module 20 is also used to obtain the task execution resource requirements and demand node type corresponding to the algorithm task to be executed; detect whether there is a computing power node that meets the task execution resource requirements, the node type is consistent with the demand node type, and is in a state to be allocated; if so, allocate the corresponding target computing power node to the algorithm task to be executed.

[0112] In a possible implementation of the present application, a shared computing power node is further provided in the power supply system, and the shared computing power node can execute the same model algorithm as the demand node; the allocation module 20 is also used to set the shared computing power node as the target computing power node corresponding to the algorithm task to be executed when the task execution resource requirements are met and the computing power nodes with the same node type as the demand node type are in an allocated state.

[0113] It should be understood that the above is only an example and does not constitute any limitation to the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any restrictions on this.

[0114] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In actual applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of this embodiment scheme, and no restrictions are imposed here.

[0115] In addition, for technical details not fully described in this embodiment, please refer to the model algorithm operation method provided in any embodiment of this application, and will not be repeated here.

[0116] In addition, 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, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. 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, method, article, or system comprising the element.

[0117] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0118] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application can essentially or in other words, the contributing part can be embodied in the form of a software product, which is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0119] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A model algorithm operation method is applied to a power supply system, wherein the power supply system is provided with interconnected demand nodes and computing nodes, wherein the demand nodes are intelligent node devices in the power supply system that can execute the model algorithm, and the computing nodes are demand nodes in an idle state; The model algorithm operation method comprises: Get the algorithm tasks to be executed; Allocate a corresponding target computing power node for the algorithm task to be executed; Control the demand node corresponding to the algorithm task to be executed to send the task execution parameters to the target computing power node, so that the target computing power node executes the algorithm task to be executed according to the task execution parameters.

2. The model algorithm operation method according to claim 1, wherein: The obtaining of the algorithm task to be executed includes: Detecting whether there is an algorithm task in a task list, wherein the task list is used to store the algorithm tasks to be executed; If it exists, the algorithm task ranked first in the task list is used as the algorithm task to be executed.

3. The model algorithm operation method according to claim 2, wherein: The task list includes a priority task list and a secondary task list, wherein the priority task list is used to store algorithm tasks with high execution priority, and the secondary task list is used to store algorithm tasks with low execution priority; The obtaining of the algorithm task to be executed includes: Detecting whether there is an algorithm task in the priority task list; If the priority task list is a blank list, detecting whether there is an algorithm task in the secondary task list; If there is an algorithm task in the secondary task list, the algorithm task ranked first in the secondary task list is used as the algorithm task to be executed.

4. The model algorithm operation method according to claim 3, wherein: Before detecting whether there is an algorithm task in the priority task list, the method further includes: After receiving the algorithm task submitted by the demand node, obtaining the delay demand value corresponding to the algorithm task; When the delay requirement value is greater than a preset delay threshold, adding the algorithm task to the secondary task list; When the delay requirement value is less than or equal to a preset delay threshold, the algorithm task is added to the priority task list.

5. The model algorithm operation method according to claim 3, wherein: Before detecting whether there is an algorithm task in the priority task list, the method further includes: Detect whether there is a waiting timeout task in the secondary task list, where the waiting timeout task is an algorithm task whose difference between the task creation time and the current time is greater than the delay requirement value; If there is a waiting timeout task, the waiting timeout task is added to the priority task list.

6. The model algorithm operation method according to claim 3, wherein: Before the detection of whether there is an algorithm task in the priority task list, the method further includes: Check whether there is a waiting timeout task in the priority task list; If there is a waiting timeout task, the sorting priority of the waiting timeout task in the priority task list is increased.

7. The model algorithm operation method according to any one of claims 1 to 6, wherein: The demand nodes include demand nodes of multiple different node types; The allocating a corresponding target computing power node to the algorithm task to be executed includes: Obtaining the task execution resource requirements and the required node type corresponding to the algorithm task to be executed; Detect whether there is a computing power node that meets the resource requirements for executing the task, whose node type is consistent with the required node type and is in a waiting-to-be-allocated state; If so, the corresponding target computing power node is assigned to the algorithm task to be executed.

8. The model algorithm operation method according to claim 7, wherein: The power supply system is also provided with a shared computing node, and the shared computing node can execute the same model algorithm as the demand node; After detecting whether there is a computing power node that meets the task execution resource requirement, whose node type is consistent with the required node type and is in an idle state, the method further includes: When the task execution resource requirement is met and all computing nodes with a node type consistent with the required node type are in an allocated state, the shared computing node is set as the target computing node corresponding to the algorithm task to be executed.

9. A model algorithm operation device, the model algorithm operation device comprising: A processor, a memory, and a model algorithm running program stored in the memory and executable on the processor, wherein the model algorithm running program, when executed by the processor, implements the steps of the model algorithm running method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, wherein a model algorithm running program is stored on the computer-readable storage medium, and when the model algorithm running program is executed, the steps of the model algorithm running method according to any one of claims 1 to 8 are implemented.

Images