Intelligent system resource scheduling method and apparatus, device, and medium

Abstract

This invention discloses a method, apparatus, device, and medium for resource scheduling in an intelligent system, relating to the field of computer technology. The method includes: acquiring first data of a first intelligent system, the first data including resource status data of the first intelligent system at a first moment; generating first resource scheduling information of the first intelligent system based on the first data using a first function; wherein the first function is a pre-constructed objective optimization function used to calculate the total contribution value of data resources, computing power resources, and model resources to the application effect of the intelligent system, and the first resource scheduling information includes the scheduling priorities and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system; and scheduling the data resources, computing power resources, and model resources in the first intelligent system respectively according to the first resource scheduling information. This invention can achieve refined collaborative scheduling of data resources, computing power resources, and model resources, which can improve the overall energy efficiency and business processing quality of the intelligent system.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, and medium for intelligent system resource scheduling. Background Technology

[0002] With the deep development and large-scale implementation of artificial intelligence, intelligent systems have been widely deployed in various business scenarios. These systems rely on their built-in model resources to provide intelligent processing and reasoning capabilities, enabling them to meet pre-defined task requirements and complete intelligent operations in various application scenarios. The effectiveness of intelligent systems highly depends on the comprehensive support of three core resources: data, computing power, and models. The rationality and effectiveness of resource scheduling have become crucial factors affecting the overall application performance of intelligent systems. Existing resource scheduling methods for intelligent systems generally suffer from unreasonable resource allocation and unclear scheduling adjustment directions, making it difficult to effectively improve the application performance of intelligent systems. Summary of the Invention

[0003] This invention provides a method, apparatus, device, and medium for resource scheduling in intelligent systems, in order to solve the problems of unreasonable resource allocation and unclear adjustment direction during resource scheduling in intelligent systems.

[0004] According to one aspect of the present invention, a resource scheduling method for an intelligent system is provided, comprising:

[0005] Acquire the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at the first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in the running state.

[0006] Based on the first data, the first resource scheduling information of the first intelligent system is generated through the first function. The first function is a pre-constructed target optimization function, with a first variable, a second variable, and a third variable as independent variables. The first function calculates the total contribution of data resources, computing power resources, and model resources to the application effect of the intelligent system. The first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the computing power resources already in use in the intelligent system, and the third variable is the business processing quality of the model resources in operation in the intelligent system. The first resource scheduling information includes the scheduling priorities and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system. The adjustment parameters indicate the direction of adjustment for each resource configuration, and the scheduling priority is positively correlated with the contribution of the resource configuration adjustments in the intelligent system to the improvement of the application effect of the intelligent system.

[0007] Based on the first resource scheduling information, the data resources, computing resources and model resources of the first intelligent system are scheduled respectively.

[0008] According to another aspect of the present invention, an intelligent system resource scheduling device is provided, comprising:

[0009] The data acquisition module is used to acquire the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at a first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing power resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in the running state.

[0010] The scheduling information generation module is used to generate first resource scheduling information of the first intelligent system based on the first data and through a first function. The first function is a pre-constructed target optimization function, with a first variable, a second variable, and a third variable as independent variables. The first function calculates the total contribution of data resources, computing power resources, and model resources to the application effect of the intelligent system. The first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the enabled computing power resources in the intelligent system, and the third variable is the business processing quality of the model resources in operation in the intelligent system. The first resource scheduling information includes the scheduling priorities and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system. The adjustment parameters indicate the adjustment direction of each resource configuration, and the scheduling priority is positively correlated with the contribution of each resource configuration adjustment in the intelligent system to improving the application effect of the intelligent system.

[0011] The resource scheduling module is used to schedule the data resources, computing resources and model resources in the first intelligent system according to the first resource scheduling information.

[0012] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the intelligent system resource scheduling method according to any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the intelligent system resource scheduling method according to any embodiment of the present invention.

[0017] According to another aspect of this application, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the intelligent system resource scheduling method described in any embodiment of this application.

[0018] The technical solution of this invention, by acquiring the first data of the first intelligent system, can accurately grasp the current data resource quality, computing power efficiency, and business processing quality of the operating model, providing comprehensive and realistic status support for resource scheduling. Based on the first data and using a target optimization function with three types of resource indicators as independent variables to generate first resource scheduling information, it can quantify the total contribution of data, computing power, and model resources to the system application effect, making the determination of scheduling priorities and adjustment parameters more objective and accurate, ensuring a high degree of matching between scheduling strategies and resource contribution levels, thereby improving the pertinence and effectiveness of resource scheduling decisions and avoiding scheduling deviations caused by subjective judgment. Scheduling data resources, computing power resources, and model resources separately based on the first resource scheduling information can avoid resource allocation imbalances and ineffective scheduling, improve the efficiency of collaborative utilization of various resources, and thus enhance the overall operational stability and processing capacity of the intelligent system. Based on the above technical solutions, by collecting resource status in multiple dimensions and driving scheduling with multi-objective optimization functions, the problems of unreasonable resource allocation and unclear adjustment direction in the resource scheduling process of intelligent systems can be solved. It can realize the fine-grained collaborative scheduling of data resources, computing power resources and model resources, thereby improving the adaptive scheduling capability of intelligent systems and improving the overall energy efficiency and business processing quality of the system.

[0019] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of an intelligent system resource scheduling method provided by an embodiment of the present invention;

[0022] Figure 2 This is a flowchart of another intelligent system resource scheduling method provided by an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the structure of an intelligent system resource scheduling device according to an embodiment of the present invention;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the intelligent system resource scheduling method of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "candidate," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Figure 1 This is a flowchart illustrating a resource scheduling method for an intelligent system provided in an embodiment of the present invention. This embodiment is applicable to scheduling various resources of an intelligent system in various intelligent system application scenarios. The method can be executed by an intelligent system resource scheduling device, which can be implemented in hardware and / or software. This intelligent system resource scheduling device can be configured in any electronic device with network communication capabilities. Figure 1 As shown in the figure, an intelligent system resource scheduling method provided by an embodiment of the present invention may include:

[0028] S110. Obtain the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at the first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in operation.

[0029] In this embodiment, the first intelligent system is the target intelligent system for which resource scheduling is to be implemented. An intelligent system refers to a system capable of completing intelligent tasks based on its built-in model resources. These model resources can refer to resources such as machine learning models that provide intelligent processing and reasoning capabilities. Completing intelligent tasks can refer to fulfilling preset task requirements in a corresponding application scenario. For example, an intelligent system can be an object or entity capable of making decisions and autonomously executing actions based on machine learning models to achieve preset goals or complete preset tasks. An intelligent system can also be an automated program that understands the user's intent and uses models or invokes tools to complete various tasks.

[0030] Resource status data can refer to a set of data reflecting the operating status and actual performance of various resources within the first intelligent system. This includes data on three core indicators: data resource quality, energy efficiency of activated computing resources, and business processing quality of the first model resources. The first model resources refer to the model resources in the intelligent system that are currently in actual operation and participate in business processing and data computation.

[0031] By real-time sensing and data collection of the resource operation status of the first intelligent system, the resource status data of the system at the first moment can be obtained. This allows for an accurate and comprehensive understanding of the actual performance and operational characteristics of the data resource quality, computing power efficiency, and business processing quality of the operation model within the first intelligent system at the current time point. This provides an objective and accurate basis for subsequent resource scheduling decisions, avoiding blindness in scheduling decisions due to a lack of or one-sided state perception.

[0032] Optionally, data resource quality is obtained by normalizing and weighted summing the multi-dimensional data quality evaluation results. These multi-dimensional data quality evaluation results are obtained by real-time monitoring of multiple quality indicators of the input data resources into the intelligent system. For structured data, multi-dimensional data quality evaluation indicators include, but are not limited to, signal-to-noise ratio, data integrity, data timeliness, sample class balance, and feature distribution offset. For unstructured data such as text, images, audio, and video, multi-dimensional data quality evaluation indicators can be obtained by calculating data resource clarity and / or information entropy values ​​using a lightweight preprocessing model.

[0033] Optionally, the energy efficiency of enabled computing resources can be obtained by taking a negative correlation mapping with the energy consumption of enabled computing resources and then normalizing it. The energy consumption of enabled computing resources can be obtained by real-time monitoring of the physical status and energy consumption indicators of the underlying computing cluster. The physical status and energy consumption indicators may include real-time power, cumulative energy consumption, GPU / TPU utilization, memory bandwidth utilization, chip temperature, and the current PUE value. For example, the energy consumption of enabled computing resources includes, but is not limited to, GPU computing power, CPU computing power, memory access energy consumption, storage read / write energy consumption, power conversion energy consumption, cooling system energy consumption, and network communication energy consumption.

[0034] Optionally, the business processing quality of the first model resource is obtained by normalizing the actual business performance evaluation results of the first model resource obtained from real-time monitoring and then weighted summing them. The business performance evaluation results of the first model resource can be obtained in real time through one or more business processing quality evaluation indicators. For example, business processing quality evaluation indicators may refer to inference latency, throughput, confidence distribution, accuracy based on validation set or online feedback, and model perplexity for large language models, etc.

[0035] S120. Based on the first data, generate the first resource scheduling information of the first intelligent system through the first function; wherein, the first function is a pre-constructed target optimization function, the first function takes a first variable, a second variable, and a third variable as independent variables, the first function is used to calculate the total contribution value of data resources, computing power resources, and model resources to the application effect of the intelligent system, the first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the computing power resources that have been activated in the intelligent system, and the third variable is the business processing quality of the model resources that are in operation in the intelligent system; the first resource scheduling information includes the scheduling priority and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system, the adjustment parameters are used to indicate the adjustment direction of each resource configuration, and the scheduling priority is positively correlated with the contribution of each resource configuration adjustment in the intelligent system to the improvement of the application effect of the intelligent system.

[0036] The first function can refer to a pre-constructed objective optimization function used to quantify the total contribution of computing resources to the system's application effect, with the first, second, and third variables as core independent variables. The application effect of an intelligent system is mainly reflected in its operational efficiency and business processing capabilities in actual business scenarios, typically through a comprehensive evaluation of performance indicators such as task processing efficiency, inference accuracy, and energy consumption.

[0037] The first variable corresponds to the quantitative representation of the data resource quality in the first intelligent system, and is a functional variable reflecting core attributes such as data resource availability and accuracy. In this embodiment, the first variable can be expressed as... ,in, For various data resource quality evaluation indicators, This indicates the number of data resource quality evaluation indicators. This represents the weighting coefficients of various data resource quality evaluation indicators, and simultaneously satisfies... and , This represents the normalization function for various data resource quality evaluation indicators. The closer the value of the first variable is to 1, the better the data resource quality and the higher the contribution of the data resource to the application effect of the intelligent system; the closer the value of the first variable is to 0, the worse the data resource quality and the lower the contribution of the data resource to the application effect of the intelligent system. Here, the contribution of the data resource to the application effect of the intelligent system refers to the data resource itself to the application effect of the intelligent system.

[0038] The second variable corresponds to the quantitative representation of the energy efficiency generated by the activated computing resources in the first intelligent system. It is a functional variable reflecting the operational efficiency and energy consumption performance of the computing resources. In this embodiment, the second variable can be expressed as: or E total This indicates the total energy consumption of the computing power already utilized in the intelligent system. This represents a normalized function for the total energy consumption of the computing power already enabled in the intelligent system. For example, E... total =E GPU算力 +E CPU算力 +E 内存访问 +E 存储读写 +E 电源转换 +E 散热系统 +E 网络通信 +E 其他 E 其他 This can be defined as the energy consumption of the intelligent system's computing resources other than GPU computing power, CPU computing power, memory access, storage read / write, power conversion, cooling system, and network communication in the application scenario. The closer the value of the second variable is to 1, the lower the energy consumption of the activated computing resources, i.e., the higher the energy efficiency, and the higher the contribution of the activated computing resources to the application effect of the intelligent system. The closer the value of the second variable is to 0, the higher the energy consumption of the activated computing resources, i.e., the lower the energy efficiency, and the lower the contribution of the activated computing resources to the application effect of the intelligent system. Here, the contribution of the activated computing resources to the application effect of the intelligent system refers to the effect of the activated computing resources themselves on the application effect of the intelligent system.

[0039] The third variable corresponds to a quantitative representation of the business processing quality of the model resources in the first intelligent system under running conditions. It is a functional variable reflecting business performance such as model processing efficiency and result accuracy. In this embodiment, the third variable can be expressed as... ,in, Evaluation indicators for the business processing quality of various model resources. This indicates the number of evaluation indicators for the quality of business processing of model resources. This represents the weighting coefficients of various model resource business processing quality evaluation indicators, and simultaneously satisfies... and , This represents the normalization function for evaluating the business processing quality of various model resources. The closer the value of the third variable is to 1, the better the business processing quality of the model resources and the higher the contribution of the model resources to the application effect of the intelligent system. The closer the value of the first variable is to 0, the worse the business processing quality of the model resources and the lower the contribution of the model resources to the application effect of the intelligent system. Here, the contribution of data resources to the application effect of the intelligent system refers to the data resources themselves to the application effect of the intelligent system.

[0040] The first resource scheduling information can refer to the information generated based on the first data and the first function, which is used to guide various resource scheduling operations within the first intelligent system. The first resource scheduling information includes the scheduling priority and adjustment parameters of data resources, computing power resources, and model resources. The scheduling priority is the scheduling priority level corresponding to the degree of contribution of each resource configuration adjustment to the improvement of the system application effect. The higher the degree of contribution, the higher the scheduling priority, and the earlier the resource configuration adjustment is carried out. The adjustment parameters are used to indicate the direction of configuration adjustment of data resources, computing power resources, and model resources, such as increasing or decreasing resources, improving or decreasing resource quality, replacing resources, etc.

[0041] Based on the first data, the data resource quality, computing power efficiency, and model business processing quality in the first data are transformed into quantitative forms of the first variable, the second variable, and the third variable, respectively. This allows the data resource quality, computing power efficiency, and model business processing quality in the first data to be analyzed under the same dimension. Based on the transformed first data and the pre-constructed first function, with the goal of optimizing the first function value, the contribution value of the three types of resources to the application effect of the intelligent system is quantitatively analyzed through the calculation logic of the first function. Then, based on the contribution value of the three types of resources to the application effect of the intelligent system, the scheduling priority and adjustment direction of each resource are determined, generating first resource scheduling information containing scheduling priority and adjustment parameters, providing clear and quantitative guidance for the scheduling execution of each resource.

[0042] The first resource scheduling information is generated based on the first data and the objective optimization function with three types of resource index data as independent variables. This can objectively and accurately quantify the contribution of each resource to the application effect of the intelligent system, so that the determination of scheduling priority and adjustment parameters is highly matched with the actual contribution of each resource. This avoids decision-making bias caused by subjective judgment in traditional scheduling, thereby improving the scientificity and rationality of resource scheduling decisions, ensuring that the resource scheduling strategy can maximize the value of resources, and improving the pertinence and effectiveness of resource scheduling.

[0043] As an optional but not limited implementation scheme, the first function includes a first function term and a second function term. The first function term is used to calculate the weighted sum of the contribution values ​​of data resources, computing power resources and model resources to the application effect of the intelligent system. The second function term is used to calculate the weighted sum of the contribution values ​​of multiple resource combinations to the application effect of the intelligent system. The resource combination is a combination of any two of the data resources, computing power resources and model resources.

[0044] In the application of intelligent systems, resources do not work independently; there are obvious nonlinear interactions between them. If the first scheduling information is determined solely by analyzing the contribution of each resource to the application effect of the intelligent system, the resource scheduling operation is likely to blindly pursue performance improvement in a single dimension, making it difficult to effectively improve the overall application effect of the intelligent system.

[0045] For example, focusing solely on improving data quality can lead to a disconnect between data and the synergy of computing power and models. Traditional data cleaning and augmentation often only focus on improving the statistical indicators of the data itself, ignoring whether the additional computing power cost required to process this data can bring a proportional improvement in model performance. Blindly pursuing high-quality data often results in the ineffective waste of computing power, thus hindering the overall application effect of the intelligent system. Similarly, focusing solely on optimizing computing power scheduling can also lead to a disconnect between computing power and the synergy of data and models. For instance, existing cluster scheduling systems mainly allocate tasks based on the utilization rate of hardware resources. However, from a practical application perspective, current cluster schedulers cannot perceive the actual value of tasks, i.e., they cannot distinguish between a high-value model gradient update and an inefficient redundant computation. Therefore, it is very easy for computing power resources to be fully utilized, but the actual model output is not good, and the overall application effect of the intelligent system is also difficult to improve effectively. Focusing solely on model efficiency optimization can lead to a disconnect between the model and the collaborative resources of computing power and data. For example, with current technology, model compression is usually completed statically before model deployment, but the model structure cannot be dynamically adjusted based on fluctuations in computing power and data quality during runtime. This often limits the overall application effect of the intelligent system and makes it difficult to achieve effective improvement.

[0046] In the application of intelligent systems, there are significant nonlinear interactions among various resources. For example, resources may exhibit synergistic effects; the combination of high-quality data and powerful computing power can produce intelligent system application effects far exceeding the simple summation of the contributions of each resource, thus generating synergistic gains. Conversely, resources may exhibit antagonistic effects; for instance, if data quality is significantly low, even a substantial increase in computing power cannot effectively improve the application effect of the intelligent system. In this case, computing power resources and data resources exhibit a negative coupling relationship, forming an antagonistic effect. Therefore, when calculating the total contribution of data resources, computing power resources, and model resources to the application effect of intelligent systems, it is necessary to fully consider the nonlinear interactions among the resources to ensure the accuracy and authenticity of the total contribution calculation.

[0047] In this embodiment, the first function includes a first function term and a second function term. The first function term calculates the independent contribution of a single resource by weighted summing of the contribution values ​​of data resources, computing power resources, and model resources, thereby achieving accurate quantification of the value of various independent resources. The second function term calculates the synergistic contribution of two resources by weighted summing of the contribution values ​​of any combination of two resources, thereby achieving a comprehensive capture of the synergistic value between resources. Through the comprehensive calculation of the first and second function terms, the total contribution value of data resources, computing power resources, model resources, and various resource combinations to the application effect of the intelligent system can be obtained, providing a quantitative basis for the subsequent generation of resource scheduling information.

[0048] The first function, through the collaborative design of the first and second function terms, can accurately quantify the independent contribution of various individual resources to the application effect of the intelligent system, and comprehensively capture the synergistic contribution value of any combination of two resources. This breaks the limitation of only considering the contribution of a single resource while ignoring the synergistic effect between resources, making the quantitative calculation of resource value more comprehensive, objective, and accurate. At the same time, through the weighted summation calculation method, weights can be allocated according to the actual impact of various resources and resource combinations on the application effect of the intelligent system in the application scenario, further improving the rationality and pertinence of the calculation results, and avoiding scheduling decision deviations caused by one-sided resource value assessment.

[0049] As an optional but not limited implementation, the second function term includes a first basis function, a second basis function, and a third basis function. The first basis function is a logarithmic growth function, used to calculate the contribution of the interaction between data resources and computing power resources to the application effect of the intelligent system. The second basis function is a saturated growth function, used to calculate the contribution of the interaction between data resources and model resources to the application effect of the intelligent system. The third basis function is a deviation penalty function, used to calculate the contribution of the matching degree between the first computing power and the second computing power to the application effect of the intelligent system. The first computing power is the computing power resource that has been enabled, and the second computing power is the computing power resource required by the model resource that is in operation.

[0050] Analysis of the interaction between data resources and computing resources reveals that when data resources are of high quality, a higher level of computing resources is required to realize their full value, creating a synergistic effect. However, once data resource quality reaches a high level, further improvements in data resource quality gradually reduce the synergistic effect on computing resources. If data resource quality is significantly low, simply increasing computing resource investment will only lead to increased energy consumption and will not effectively improve the application effect of the intelligent system. In this case, the intelligent system should reduce computing resource investment to save costs. Therefore, the synergistic effect between data resources and computing resources exhibits a logarithmic growth relationship.

[0051] In this embodiment, the contribution of the interaction between data resources and computing power resources to the application effect of the intelligent system is calculated using a first basis function. Specifically, the first basis function can be expressed as follows: ,in, The first variable is used to represent the quality of data resources. The second variable represents the energy efficiency generated by the activated computing resources. As can be seen from the first basis function, when... When, that is, when the quality of data resources is extremely poor, This means that no matter how the computing resources are adjusted, no synergistic gains can be generated.

[0052] Analysis of the interaction between data resources and model resources reveals that complex models require high-quality data resources to fully realize their application performance, while simple models are relatively less sensitive to data quality. For most model resources, if the data quality is low, the model resources typically struggle to achieve optimal performance. Improving data resource quality can enhance the business processing performance of model resources, but all types of model resources have inherent performance limits. Even with continuous improvement in data resource quality, the business processing performance of model resources cannot be increased indefinitely. Therefore, the synergistic effect between data resources and model resources exhibits a saturated growth relationship.

[0053] In this embodiment, the contribution of the interaction between data resources and model resources to the application effect of the intelligent system is calculated using a second basis function. Specifically, the second basis function can be expressed as follows: ,in, , The first variable is used to represent the quality of data resources. The third variable represents the quality of business processing of model resources in operation. As can be seen from the second basis function, as the quality of data resources improves, the collaborative gain indicated by the value of the second basis function rises rapidly. When the quality of data resources improves to a certain extent, diminishing marginal benefits will occur.

[0054] Analysis of the interaction between computing resources and model resources reveals that different model resources have varying utilization rates of computing resources. For example, model resources with complex structures and a large number of parameters have higher computing resource requirements. If the actual computing resources used exceed the computing power required by the running model resources, it will result in computing resource redundancy and energy waste. Conversely, if the actual computing resources used are less than the computing power required by the running model resources, it will lead to insufficient computing resource supply, causing model resource lag or even malfunction. Therefore, any mismatch between computing resources and model resources will reduce the overall application effect of the intelligent system. Thus, the synergistic effect between computing resources and model resources exhibits a deviation penalty relationship; that is, when there is a configuration deviation between the two, it will negatively impact the overall application effect of the intelligent system, requiring the application of corresponding penalty constraints.

[0055] In this embodiment, the contribution of the interaction between computing resources and model resources to the application effect of the intelligent system is calculated using a third basis function. Specifically, the third basis function can be expressed as follows: ,in, The second variable represents the energy efficiency generated by the activated computing resources. This represents the computing resources required to represent the model resources in a running state. As a preset constant, it can be obtained in advance through historical experience or model calculation and iteration. According to the third basis function, the cooperative gain is maximized when the computing power required by the activated computing power resources matches the computing power required by the model resources in operation. When the activated computing power resources are redundant or insufficient, the value of the third basis function is negative, which means that the reduction in the application effect of the intelligent system due to the mismatch between the supply and demand of computing power resources is punished.

[0056] Based on the aforementioned first basis function, second basis function, and third basis function, the first function can be expressed as: ,in, These are the first, second, and third variables mentioned above, respectively; For the first function term, These are the weights of the contributions of data resources, computing power resources, and model resources to the application effect of the intelligent system, determined according to the specific business scenario of the intelligent system. For example, in a business scenario with the goal of energy conservation and carbon reduction, Maximum; in medical-related business scenarios, maximum; For the second function term, These are the aforementioned first basis functions, second basis functions, and third basis functions, respectively. The weights for the contribution values ​​of each resource combination to the application effect of the intelligent system are also determined based on the specific business scenarios of the intelligent system. This is a constant disturbance term.

[0057] In this embodiment, the second function term sets three basis functions with differentiated functions that are adapted to the interaction rules of corresponding resource combinations. These functions accurately calculate the synergistic characteristics of different resource combinations, breaking the limitation that a single function form cannot adapt to the interaction rules of various resource combinations. This makes the calculation of the synergistic contribution value of resource combinations more accurate and realistic. By weighted summing the calculation results of the three basis functions, the second function term's calculation of the synergistic contribution value of resource combinations is more comprehensive and objective. This further improves the scientificity and reliability of the calculation result of the total contribution value of the first function, reduces scheduling deviations caused by inaccurate assessment of the synergistic contribution of resource combinations, ensures the refinement and synergy of resource scheduling in the intelligent system, and improves the overall system operation effect and resource utilization efficiency.

[0058] S130. Based on the first resource scheduling information, schedule the data resources, computing power resources and model resources in the first intelligent system respectively.

[0059] Based on the generated first resource scheduling information, and according to the order of scheduling priority and the adjustment direction indicated by the adjustment parameters, targeted scheduling operations are carried out on the data resources, computing resources, and model resources in the first intelligent system to achieve refined and targeted management and configuration optimization of the three types of resources. The scheduling operation refers to the process of adjusting and optimizing the configuration, allocation, and operating status of resources. The specific scheduling operation for each resource can be determined according to the actual application scenario of the intelligent system. In this embodiment, the selection method for the specific scheduling operation is not limited. For example, it can be determined according to a preset scheduling operation table, by a pre-trained action selection model, or by technicians based on resource scheduling information. For example, the scheduling operation can include specific operations such as enabling, disabling, expanding, and reducing resources.

[0060] Optionally, the intelligent system includes a computing resource scheduler, a data resource scheduler, and a model resource scheduler. The computing resource scheduler interfaces with the underlying resource management system to perform computing resource scheduling operations such as node scaling, voltage and frequency adjustment, and task migration. The data resource scheduler interfaces with the data pipeline to perform data resource scheduling operations such as data filtering, enhancement, sampling rate adjustment, or enabling specific cleaning algorithms. The model resource scheduler interfaces with the model service grid to perform model resource scheduling operations such as model hot switching, dynamic routing, parameter fine-tuning, or enabling early termination mechanisms.

[0061] The technical solution of this invention, by acquiring the first data of the first intelligent system, can accurately grasp the current data resource quality, computing power efficiency, and business processing quality of the operating model, providing comprehensive and realistic status support for resource scheduling. Based on the first data and using a target optimization function with three types of resource indicators as independent variables to generate first resource scheduling information, it can quantify the total contribution of data, computing power, and model resources to the system application effect, making the determination of scheduling priorities and adjustment parameters more objective and accurate, ensuring a high degree of matching between scheduling strategies and resource contribution levels, thereby improving the pertinence and effectiveness of resource scheduling decisions and avoiding scheduling deviations caused by subjective judgment. Scheduling data resources, computing power resources, and model resources separately based on the first resource scheduling information can avoid resource allocation imbalances and ineffective scheduling, improve the efficiency of collaborative utilization of various resources, and thus enhance the overall operational stability and processing capacity of the intelligent system. Based on the above technical solutions, by collecting resource status in multiple dimensions and driving scheduling with multi-objective optimization functions, the problems of unreasonable resource allocation and unclear adjustment direction in the resource scheduling process of intelligent systems can be solved. It can realize the fine-grained collaborative scheduling of data resources, computing power resources and model resources, thereby improving the adaptive scheduling capability of intelligent systems and improving the overall energy efficiency and business processing quality of the system.

[0062] Figure 2 This is a flowchart of another intelligent system resource scheduling method provided by an embodiment of the present invention. This embodiment further refines the process in the above embodiment of generating first resource scheduling information of a first intelligent system based on first data through a first function. For example... Figure 2 As shown in the figure, another intelligent system resource scheduling method provided by the embodiments of the present invention may include:

[0063] S210. Obtain the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at the first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing power resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in the running state.

[0064] Optionally, before performing resource scheduling on the intelligent system, the business requirements of the intelligent system, the adjustable range of various resources, and the initial configuration of each resource are determined. The business requirements are used to indicate the expected effect of the intelligent system performing resource scheduling. For example, business requirements may include service level agreement requirements, such as a business processing accuracy greater than 95% and a business processing latency less than 50 milliseconds. The adjustable range of various resources is used to constrain resource adjustment services, such as the upper limit of the hardware resource pool and energy consumption budget. The initial configuration of each resource is used to provide initial resource status data for resource scheduling operations, such as setting the initial model, accessing data source type, data volume, and data quality.

[0065] In this embodiment, the parameter values ​​in the first function can be set according to the business needs of the intelligent system, the adjustable range of various resources, and the initial configuration of each resource, combined with historical experience, so as to ensure that the scheduling process is based on a better initial resource state and that each resource configuration is scheduled to the optimal state as soon as possible.

[0066] S220. Based on the first data and the first function, determine the first scheduling priority of each resource in the first intelligent system.

[0067] The first scheduling priority refers to the initial scheduling order of the three types of resources in an intelligent system, calculated based on the first data and the first function during a complete scheduling process. A complete scheduling process for the three resources in an intelligent system refers to the overall scheduling flow in which scheduling operations are performed on data resources, computing resources, and model resources. Based on the collected first data and a pre-constructed first function, quantitative calculations are performed. According to the calculated contribution of each type of resource to the system's application effect, the first scheduling priority of each of the data resources, computing resources, and model resources is determined, thus achieving a preliminary division of the scheduling order of each type of resource.

[0068] As an optional but not limited implementation, the first scheduling priority of each resource in the first intelligent system is determined based on the first data and the first function, including:

[0069] Based on the first data and the first function, the first parameter, the second parameter, and the third parameter are determined. The first parameter is the partial derivative of the first function with respect to the first variable. The second parameter is the partial derivative of the first function with respect to the second variable. The third parameter is the gain value of the second model resource. The second model resource is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency indicated in the first data. The gain value is the difference between the first function value after predicting the model switch and the current first function value. The current first function value is calculated based on the first data.

[0070] Based on the first parameter, the second parameter, the third parameter, and the preset first weight combination, the first scheduling priority of each resource in the first intelligent system is determined. The first scheduling priority of data resources is the product of the absolute value of the first parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of computing power resources is the product of the absolute value of the second parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of model resources is the product of the third parameter and the corresponding weight in the preset first weight combination.

[0071] Both data resource quality and the energy efficiency of activated computing resources are continuous variables. That is, the first and second variables in the first function are both continuous variables. By taking the partial derivatives of the first function with respect to the first and second variables, we can quantify the direction and degree of influence of each unit adjustment of the first and second variables on the total contribution value represented by the first function value. By calculating the first and second parameters, we can accurately capture the magnitude of the impact of changes in data resources and computing resources on the system application effect, quantifying the correlation between resource status changes and system application effect. This avoids the one-sidedness of judging resource importance solely through qualitative analysis and improves the rationality and targeting of scheduling priority allocation.

[0072] In this embodiment, based on the first data and the first function, the partial derivative of the first function with respect to the first variable under the first data condition is calculated and denoted as the first parameter. The first parameter indicates the direction and degree of influence of each unit adjustment of the first variable on the total contribution value indicated by the first function value under the current resource configuration state indicated by the first data, that is, the direction and degree of influence of data resource quality on the overall application effect of the intelligent system. Similarly, based on the first data and the first function, the partial derivative of the first function with respect to the second variable under the first data condition is calculated and denoted as the second parameter. The second parameter indicates the direction and degree of influence of each unit adjustment of the second variable on the total contribution value indicated by the first function value under the current resource configuration state indicated by the first data, that is, the direction and degree of influence of each unit adjustment of computing power resource efficiency on the overall application effect of the intelligent system. The absolute values ​​of the first and second parameters respectively represent the contribution of the adjustment of the corresponding variable values ​​of data resources and computing power resources to the improvement of the application effect of the intelligent system.

[0073] Internal structural optimization of the same model resource typically requires stopping the model resource's operation. To ensure the continuity of the intelligent system's operation, this embodiment adjusts model resources through model resource switching or online adjustment. That is, during the operation of the intelligent system, model resource adjustments are achieved through methods such as hot switching, early termination mechanisms, or online parameter fine-tuning. For example, switching to a model resource that better matches the business logic can directly improve the quality of business processing.

[0074] In this embodiment, the adjustment of model resources is a discrete operation, and the corresponding improvement in business processing quality is not continuous. Therefore, the third variable in the first function is considered a discrete variable. By traversing the model resources whose data resource quality and computing power efficiency can support operation as indicated by the first data, the gain value after switching to each model resource is calculated, and the maximum gain value is selected as the third parameter. The sign of the gain value can characterize the improvement or reduction of the application effect of the intelligent system after model switching, and the magnitude of the gain value can characterize the magnitude of the impact on the application effect of the intelligent system. The third parameter can represent the maximum contribution that model resource adjustment can bring to the improvement of the application effect of the intelligent system under the current resource configuration state indicated by the first data.

[0075] The preset first weight combination is a pre-defined set of weights, which includes three weights corresponding to data resources, computing resources, and model resources, respectively. These weights are used to balance the influence of the first, second, and third parameters on the scheduling priority. The size of each weight is pre-configured according to the actual application scenario of the intelligent system. For example, it can be configured based on factors such as the adjustable range of each resource and the ease of adjustment of each resource in the actual application scenario.

[0076] In this embodiment, the first scheduling priority can refer to the initial scheduling order of data resources, computing power resources, and model resources in the first intelligent system during a complete scheduling process of the three resources in the intelligent system, determined according to the product of the first parameter, the second parameter, the third parameter, and the corresponding weights. Specifically, the first scheduling priority of data resources can refer to the product of the absolute value of the first parameter and the weight of the corresponding data resource in the preset first weight combination, used to characterize the scheduling priority of data resources; the first scheduling priority of computing power resources can refer to the product of the absolute value of the second parameter and the weight of the corresponding computing power resource in the preset first weight combination, used to characterize the scheduling priority of computing power resources; and the first scheduling priority of model resources can refer to the product of the third parameter and the weight of the corresponding model resource in the preset first weight combination, used to characterize the scheduling priority of model resources.

[0077] By using the first, second, and third parameters as core contribution quantification indicators for various resources, and combining them with a preset first weight combination, the first scheduling priority values ​​for data resources, computing power resources, and model resources are determined. By comparing the magnitudes of the three priority values, the final order of scheduling for various resources is determined, ensuring that the scheduling priority can comprehensively reflect the contribution and influence weight of various resources to the system application effect.

[0078] S230. Based on the first scheduling priority of each resource in the first intelligent system, determine the first resource, which is the resource with the highest first scheduling priority in the first intelligent system.

[0079] By comparing and selecting resources based on the first scheduling priority, the resource with the highest scheduling priority value among data resources, computing resources, and model resources is selected as the first resource. This can quickly identify the resource that contributes the most to improving the application effect of the intelligent system among the three types of resources, giving the resource scheduling behavior a clear direction and prioritizing the adjustment of resources that can bring greater effect improvement. This can improve the focus and decision-making efficiency of scheduling execution.

[0080] As an optional but not limited implementation, the method of generating first resource scheduling information for the first intelligent system through a first function based on first data further includes:

[0081] Based on the first parameter, the second parameter, and the preset step size, the first adjustment parameter and the second adjustment parameter are determined. The first adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the data resource, and the second adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the computing power resource. The first parameter is the partial derivative value of the first function with respect to the first variable, and the second parameter is the partial derivative value of the first function with respect to the second variable.

[0082] Based on the first data and the first function, a third adjustment parameter is determined. The third adjustment parameter is used to indicate the second model resource. The second model resource is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency in the first data. The gain value is the difference between the first function value after predicting the switching of model resources and the current first function value. The current first function value is calculated based on the first data.

[0083] Based on the first data and the first function, a first parameter and a second parameter are calculated. The first parameter indicates the direction and degree of influence of each unit adjustment of the first variable on the total contribution value indicated by the first function value under the current resource configuration indicated by the first data, i.e., the direction and degree of influence of data resource quality on the overall application effect of the intelligent system. The second parameter indicates the direction and degree of influence of each unit adjustment of the second variable on the total contribution value indicated by the first function value under the current resource configuration indicated by the first data, i.e., the direction and degree of influence of each unit adjustment of computing power resource efficiency on the overall application effect of the intelligent system. The absolute values ​​of the first and second parameters represent the contribution of data resource and computing power resource configuration adjustments to the improvement of the application effect of the intelligent system, respectively; the positive and negative values ​​of the first and second parameters represent the direction of data resource and computing power resource configuration adjustments, respectively. A negative value indicates a decrease in the variable value corresponding to the resource, and a positive value indicates an increase in the variable value corresponding to the resource. For example, a positive first parameter indicates that the application effect of the intelligent system needs to be improved by enhancing the quality of data resources; a negative first parameter indicates that there may be over-cleaning of data resources, and the application effect of the intelligent system can be indirectly improved by reducing data processing; a positive second parameter indicates that more computing power resources need to be invested to improve the application effect of the intelligent system; a negative second parameter indicates that there may be redundant computing power, and computing power investment needs to be reduced to indirectly improve the application effect of the intelligent system by reducing energy consumption.

[0084] By calculating the first and second parameters, the impact of changes in the state of data resources and computing power resources on the system's application effect can be accurately quantified. This breaks the limitation of relying solely on qualitative judgments of resource adjustment direction and provides objective and reliable quantitative support for the scientific determination of subsequent data resource and computing power resource adjustment parameters. This ensures that the adjustment parameters can accurately match the impact of resource state changes on the system's application effect.

[0085] In this embodiment, the preset step size refers to a pre-set benchmark value used to control the adjustment range of resource configuration, preventing system instability caused by excessively large or small resource adjustment ranges. The preset step size can be pre-set according to the actual application scenario of the intelligent system. For example, a smaller preset step size can be set for scenarios where the application effect of the intelligent system fluctuates more significantly to ensure the stability and reliability of scheduling. Optionally, during the resource scheduling process of the intelligent system, the preset step size can be decayed according to a preset ratio to gradually optimize the adjustment range of resource configuration and avoid instability in scheduling effect due to excessively large step sizes. Examples include exponential decay, decay using different preset ratios in stages, and continuous linear decay.

[0086] In this embodiment, the first adjustment parameter can be expressed as: ,in As the first parameter, With a preset step size, the sign of the first adjustment parameter indicates the direction of adjustment for the first variable corresponding to the data resource; a positive value indicates improved data resource quality, and a negative value indicates decreased data resource quality. The absolute value of the first adjustment parameter indicates the degree of adjustment for the first variable corresponding to the data resource, i.e., the adjustment magnitude of the first variable in each adjustment. Adjusting the first variable corresponding to the data resource according to the first scheduling parameter can be expressed as follows: ,in This represents the first variable after adjustment. The second adjustment parameter can be expressed as... ,in For the second parameter, With a preset step size, the sign of the second adjustment parameter indicates the direction of adjustment for the second variable corresponding to the data resource; a positive value indicates increased computing resource investment, and a negative value indicates decreased computing resource investment. The absolute value of the second adjustment parameter indicates the degree of adjustment of the second variable corresponding to the computing resource, i.e., the adjustment magnitude of the second variable in each adjustment. Adjusting the second variable corresponding to the computing resource according to the second scheduling parameter can be expressed as follows: ,in This represents the adjusted second variable.

[0087] By combining the first parameter, the second parameter, and the preset step size to determine the first and second adjustment parameters, the direction and degree of adjustment of data resources and computing power resources can be precisely matched to the impact of resource status changes on the system application effect. At the same time, by controlling the adjustment range through the preset step size, it is possible to avoid the system operation fluctuation caused by excessive resource adjustment range, or the optimization effect not being achieved due to insufficient adjustment range. This ensures that the adjustment operation of data resources and computing power resources is scientific and controllable, and improves the accuracy and effectiveness of resource adjustment.

[0088] Based on the first data and the first function, the system first determines the third model resource that can support operation by combining the data resource quality and computing power efficiency status in the first data. Then, for each model resource in the third model resource, the system predicts the value of the first function after the switch and calculates the difference between the function value and the current first function value to obtain the gain value corresponding to each type of model resource. The model resource with the largest gain value is selected as the second model resource. Finally, the relevant information indicating the second model resource is determined as the third adjustment parameter. The prediction of the first function value after the switch can be achieved by predicting the business processing quality of the model resource through a performance prediction model, by fitting historical experience data, or by simulating the operation of the model resource. This embodiment does not limit the method of predicting the first function value after the switch.

[0089] By determining the third adjustment parameter, the optimal target for model resource adjustment can be accurately identified, the switching direction of model resources can be clarified, and the blindness of model resource adjustment can be avoided. At the same time, using the gain value as the screening criterion, it is ensured that the switched model resources can maximize the system application effect, so that the adjustment of model resources is highly consistent with the system optimization goal, further improving the pertinence and effectiveness of resource scheduling, and giving full play to the role of model resources in improving the application effect of intelligent system.

[0090] S240. Based on the first data and the first function, adjust the value of the variable corresponding to the first resource to obtain the second data. The second data includes the resource status data of the first intelligent system at the second moment. The second moment is the moment when the value of the variable corresponding to the first resource is adjusted.

[0091] Based on the first data and the calculation results of the first function, the configuration adjustment operation is performed on the variable values ​​corresponding to the determined first resource. At the second moment after the adjustment is completed, the system resource status is re-determined, obtaining the updated second data for each resource status, providing the latest resource status basis for subsequent scheduling. The variable values ​​corresponding to the first resource are used to indicate the status of the first resource; for example, the variable value corresponding to data resources is the first variable value, the variable value corresponding to computing power resources is the second variable value, and the variable value corresponding to model resources is the third variable value. Since only the variable values ​​corresponding to the resources are adjusted, and the resource itself has not yet changed its configuration, the intelligent system still operates in the state indicated by the first data.

[0092] Optionally, adjusting the variable value corresponding to the first resource based on the first data and the first function includes: obtaining the scheduling parameters corresponding to the first resource; and adjusting the variable value corresponding to the first resource based on the scheduling parameters. For example, the first resource is a data resource; a first parameter is calculated based on the first data and the first function, where the first parameter is the partial derivative of the first function with respect to the first variable; a first adjustment parameter is determined based on the first parameter and a preset step size, where the first adjustment parameter indicates the direction and degree of adjustment of the variable value corresponding to the data resource; and the variable value corresponding to the first resource is adjusted based on the first adjustment parameter.

[0093] By adjusting the variable values ​​corresponding to the first resource and obtaining the updated second data, it is possible to prioritize the optimization of core resources while synchronizing resource status changes in real time. This ensures that subsequent scheduling calculations are always based on the latest intelligent system status, thereby improving the dynamic adaptability of the scheduling process.

[0094] S250. Based on the second data and the first function, determine the second scheduling priority of each remaining resource other than the first resource.

[0095] Based on the second data after the first resource adjustment, a new quantification calculation is performed using the first function. Only for the two remaining resource categories excluding the first resource, the corresponding second scheduling priority is determined, completing the order update and partitioning of resources that have not yet been scheduled. Re-determining the scheduling priority of the remaining resources based on the updated second data ensures that the scheduling order is updated in accordance with the actual operating state after resource adjustment, avoiding scheduling decision lags caused by using the initial priority, and guaranteeing the rationality and timeliness of subsequent resource adjustments.

[0096] S260. Determine the second resource based on the second scheduling priority. The second resource is the resource with the highest second scheduling priority in the first intelligent system.

[0097] Based on the second scheduling priority, the resource with the highest scheduling priority among the remaining unadjusted resources is selected as the second resource. Determining the second resource through the second scheduling priority allows for further optimization of resources that contribute the second highest to improving system application performance after the core resources have been adjusted. This ensures that the scheduling process follows the principle of gradient optimization, further enhancing the systematic nature of resource scheduling.

[0098] S270. Based on the second data and the first function, adjust the value of the variable corresponding to the second resource to obtain the third data. The third data includes the resource status data of the first intelligent system at the third moment. The third moment is the moment when the adjustment of the variable corresponding to the second resource is completed.

[0099] Based on the second data and the first function, the variable values ​​corresponding to the second resource are adjusted. At the third moment after the adjustment is completed, the latest resource status data is obtained, providing updated resource status data for the scheduling adjustment of the last type of resource. By making targeted configuration adjustments to the second resource and obtaining the third data, the system resource configuration structure can be continuously optimized, while resource status information is updated in real time, providing reliable data support for the accurate adjustment of the last type of resource and ensuring the continuity of the scheduling process.

[0100] Optionally, adjusting the variable value corresponding to the second resource based on the second data and the first function includes: obtaining the scheduling parameters corresponding to the second resource; and adjusting the variable value corresponding to the second resource based on the scheduling parameters. For example, the second resource is a computing resource. A second parameter is calculated based on the second data and the first function, where the second parameter is the partial derivative of the first function with respect to the second variable. A second adjustment parameter is determined based on the second parameter and a preset step size. The second adjustment parameter indicates the direction and degree of adjustment of the variable value corresponding to the computing resource. The variable value corresponding to the second resource is then adjusted based on the second adjustment parameter.

[0101] S280. Based on the third data and the first function, adjust the variable value corresponding to the third resource. The third resource is the resource in the first intelligent system other than the first resource and the second resource.

[0102] Based on the third data obtained after adjusting the first two types of resources, and combined with the calculation completed by the first function, the corresponding variable values ​​of the third resource are adjusted. This completes a full scheduling process for the three resources in the intelligent system. The third resource is the only remaining resource among the three that has not yet had its corresponding variable values ​​adjusted. By adjusting the corresponding variables of the third resource based on the latest third data, complete scheduling coverage of all system resources can be achieved, avoiding resource scheduling omissions. This allows for targeted optimization of data resources, computing power resources, and model resources, comprehensively improving the integrity of resource scheduling and the overall system performance.

[0103] Optionally, the variable value corresponding to the third resource is adjusted based on the third data and the first function, including: obtaining the scheduling parameters corresponding to the third resource; and adjusting the variable value corresponding to the third resource based on the scheduling parameters. For example, the third resource is a model resource. A third parameter is calculated based on the third data and the first function. The third parameter is the gain value of the fourth model. The fourth model resource is the model resource with the largest gain value among the fifth model resources. The fifth model resource is the model resource that can be supported by the data resource quality and computing power efficiency indicated in the third data. The gain value is the difference between the first function value after predicting the model switch and the current first function value, which is calculated based on the third data. A third adjustment parameter is determined based on the third parameter and a preset step size. The third adjustment parameter is used to indicate the fourth model resource. The variable value corresponding to the third resource is adjusted based on the third adjustment parameter.

[0104] For example, suppose that for an intelligent system, the scheduling priority of data resources in the system is calculated to be 6.3, the scheduling priority of computing resources is 8.1, and the scheduling priority of model resources is 2.0. Therefore, it is determined that computing resources should be adjusted first. After adjusting the variable values ​​corresponding to the computing resources according to a preset step size, the status data of each resource in the intelligent system is reacquired. Based on the latest status data of each resource in the intelligent system, the scheduling priority of model resources is calculated to be 5.4, and the scheduling priority of data resources is 4.2. Therefore, it is determined that after adjusting computing resources, model resources should be adjusted first. After adjusting the variable values ​​corresponding to model resources according to a preset step size, the status data of each resource in the intelligent system is reacquired. Based on the latest status data of each resource in the intelligent system, the data resources that have not yet been adjusted are adjusted. In this example, the scheduling order of a complete scheduling process for the three resources in the intelligent system is computing resources, model resources, and data resources. Since adjusting one resource will affect the status data of other resources, after each complete adjustment of a resource, it is necessary to reacquire the resource loading information of the intelligent system to ensure that each adjustment is based on the local optimal direction of the current resource status of the intelligent system, and to avoid the failure of resource adjustment due to changes in the system resource status.

[0105] S290. Based on the first resource scheduling information, schedule the data resources, computing power resources and model resources in the first intelligent system respectively.

[0106] Optionally, in this embodiment, it is permissible to schedule the first resource in the first intelligent system immediately after determining the first resource, to schedule the second resource in the first intelligent system immediately after determining the second resource, and to schedule the third resource in the first intelligent system immediately after determining the third resource; it is also permissible to schedule the first resource, the second resource, and the third resource in the first intelligent system sequentially at once after determining the first resource, the second resource, and the third resource.

[0107] As an optional but not limited implementation, after scheduling the data resources, computing resources, and model resources in the first intelligent system according to the first resource scheduling information, the intelligent system resource scheduling method provided in this embodiment of the invention may further include:

[0108] Based on the first data, the value of the first function at the first moment is calculated using the first function.

[0109] The fourth data is obtained, which includes the resource status data of the first intelligent system at the fourth moment. The fourth moment is the moment after the scheduling of each resource in the first intelligent system is completed according to the first resource scheduling information.

[0110] Based on the fourth data, the value of the first function at the fourth time point is calculated using the first function.

[0111] Calculate the gain value of this resource scheduling based on the first function value at the first time step and the first function value at the fourth time step;

[0112] If the gain value of this resource scheduling is not greater than the preset gain threshold, the cumulative number of invalid resource scheduling counts is incremented by one;

[0113] If the gain value of this resource scheduling is greater than the preset gain threshold, the cumulative number of invalid resource scheduling counts is reset to zero.

[0114] If the cumulative number of invalid resource schedulings is not less than the preset scheduling number threshold, and the preset step size is less than the preset minimum step size, resource scheduling will stop. The preset step size is used to indicate the degree of configuration adjustment for each resource scheduling.

[0115] If the cumulative number of invalid resource scheduling attempts is less than the preset scheduling attempt threshold, the preset step size is decayed according to a preset ratio, and the system returns to reacquire the resource status data of the first intelligent system for the next resource scheduling attempt.

[0116] Based on the initial data collected before resource scheduling, this data is substituted into a pre-constructed function for quantitative calculation, yielding the first function value at the first moment. This value serves as a benchmark quantitative indicator of the system's application effect before resource scheduling. By calculating the first function value at the first moment, a clear and objective benchmark can be established for evaluating the resource scheduling effect. This avoids the inability to accurately determine the effectiveness of scheduling operations due to a lack of benchmark values, providing a reliable basis for subsequent gain calculations and scheduling effect assessments, and ensuring the scientific nature of the scheduling effect evaluation.

[0117] After all resource scheduling operations are completed, resource status data of the first intelligent system is collected at the fourth time step to obtain the fourth data. This data comprehensively captures the latest operating status of data resources, computing power resources, and model resources within the system after scheduling. By acquiring the fourth data, the system resource status after resource scheduling can be grasped in real time, ensuring that subsequent scheduling effect evaluation is based on the actual operating situation after scheduling. This avoids distortion of evaluation results due to lag in status information and provides objective and comprehensive data support for accurately calculating scheduling gains and judging scheduling effectiveness.

[0118] The fourth data point is substituted into the first function for quantification to obtain the first function value at the fourth time step. This value serves as a quantitative indicator of the system's application effect after resource scheduling, and is compared with the first function value at the first time step. By calculating the first function value at the fourth time step, the actual state of the system's application effect after resource scheduling can be accurately quantified, forming a clear comparison with the baseline value before scheduling. This provides a quantitative basis for objectively evaluating the actual effect of resource scheduling, ensuring that the evaluation of scheduling effect is operable and accurate.

[0119] Based on the first function values ​​at the first and fourth time points, the gain value of this resource scheduling is obtained by calculating the difference between the two values. This gain value directly reflects the degree to which this scheduling operation improves the system's application performance. By calculating the gain value of this resource scheduling, the effect of the scheduling operation can be quantified, allowing for a clear and intuitive judgment as to whether the scheduling has achieved the goal of improving the system's application performance. This avoids biases in the qualitative judgment of the scheduling effect and provides a clear decision-making basis for adjusting subsequent scheduling strategies.

[0120] The gain value of this resource scheduling is compared with a preset gain threshold. If the gain value does not exceed the preset gain threshold, it indicates that the scheduling has not achieved the expected system optimization effect and is judged as invalid scheduling. In this case, the cumulative invalid resource scheduling count is incremented by one to track the cumulative status of invalid scheduling. By counting invalid scheduling, the effectiveness of resource scheduling can be tracked in real time, problems with the scheduling strategy can be identified in a timely manner, and data support can be provided for subsequent adjustments to the scheduling step size and optimization of the scheduling strategy, avoiding resource waste and decreased system operating efficiency caused by continuous execution of invalid scheduling. If the gain value exceeds the preset gain threshold, it indicates that the scheduling has achieved the expected system optimization effect and is judged as effective scheduling. In this case, the cumulative invalid resource scheduling count is reset to zero and the counting starts again, ensuring that the number of invalid scheduling accurately reflects the recent scheduling effect. By resetting the number of invalid scheduling to zero after effective scheduling, changes in recent scheduling effects can be accurately tracked, avoiding the accumulation of invalid scheduling counts from misleading subsequent scheduling strategy adjustments. At the same time, the effect of effective scheduling is clearly defined, providing a basis for maintaining the current reasonable scheduling strategy and ensuring the dynamic optimization of the scheduling strategy.

[0121] By using a dual check of the cumulative number of invalid resource scheduling attempts and a preset step size, the system determines whether to continue the next resource scheduling operation. If the cumulative number of invalid scheduling attempts reaches or exceeds the preset threshold, and the current preset step size is already smaller than the preset minimum step size, it indicates that adjusting the step size will no longer improve the scheduling effect, and continuing the scheduling operation will only waste system resources. Therefore, the resource scheduling operation is terminated. This dual check effectively avoids the continuous execution of invalid scheduling, reduces the waste of system resources, and prevents scheduling inefficiency caused by excessively small step sizes. This ensures the rationality and economy of resource scheduling operations and avoids the impact of over-scheduling on system stability.

[0122] When the cumulative number of invalid resource scheduling attempts has not reached the preset threshold, it indicates that there is still an opportunity to improve the scheduling effect by adjusting the step size. At this point, the preset step size is decayed according to a preset ratio, reducing the adjustment range of subsequent resource scheduling. Then, the process returns to the initial step, the state data of the intelligent system is collected again, and the next resource scheduling cycle is started, realizing dynamic optimization of the scheduling strategy. By decaying the preset step size proportionally and starting the next scheduling cycle, the adjustment range of resource allocation can be gradually optimized, avoiding the instability of scheduling effect due to excessively large step sizes. At the same time, by continuously trying to optimize the system effect through cyclical scheduling, it is ensured that resource scheduling can adapt to changes in system state, improving the flexibility and effectiveness of the scheduling strategy, and ultimately achieving continuous improvement in the system application effect.

[0123] For example, taking a smart medical image analysis system as an example, in the scenario of smart medical image analysis, the amount of medical CT / MRI image data is huge, and the requirements for diagnostic accuracy are extremely high. The computing power at the hospital edge is limited, and energy consumption and heat dissipation need to be strictly controlled. Therefore, for the first function, the business processing quality weight of model resources is the highest, the data resource quality weight is the second highest, and the energy efficiency weight of computing power resources is the lowest. In addition, according to the characteristics of the application scenario, the synergistic gain of data resources and model resources can be utilized to improve the overall application effect of the system. For example, a doctor submits a batch of lung CT scans, requiring a high-precision analysis mode. Resource status data is obtained. Since the resource status data indicates that some images have motion artifacts caused by the patient's breathing, i.e., the data resource quality is low, the scheduling priority and adjustment parameters of each resource are obtained, indicating that the data resource quality should be improved first. When the data resource quality is improved to a high standard, the resource status data is obtained again, and through analysis and calculation, the model resources are adjusted to improve system performance by adopting a high-precision model. After obtaining the resource status data again, the adjustment direction and magnitude of computing power resources are determined. Finally, adjustments were made sequentially based on the direction and magnitude of each resource adjustment. The adjustment process was as follows: Input data was processed using a motion artifact correction algorithm, which temporarily increased computational power consumption but significantly improved data quality, resulting in the optimal system performance improvement for this step. A large cloud-based model was then used to process the enhanced data. Given that the edge computing layer could not support the simultaneous operation of the large model and the enhancement algorithm, the system, through network-based collaboration, scheduled data enhancement and analytical inference tasks to a dedicated medical cloud cluster for execution, with the edge computing layer only responsible for result display. Through this complete scheduling process of the three resources in the intelligent system, some transmission and cloud computing power consumption were sacrificed in exchange for a significant improvement in the accuracy of the analysis, maximizing the first function value and enhancing the overall application effect of the intelligent medical image analysis system.

[0124] For example, taking a high-frequency trading risk control system in finance as an example, in financial trading scenarios, the flow rate of financial trading data is extremely fast, even reaching the microsecond level, making it extremely sensitive to the business processing efficiency of model resources. Computing power consumption is also a major cost consideration, but it is secondary to latency. Furthermore, due to market fluctuations, financial data may contain a large amount of noise. Therefore, for the first function, the business processing quality of model resources has the highest weight, the energy efficiency weight of computing power resources is the second highest, and the quality weight of data resources is the lowest. In addition, based on the characteristics of the application scenario, the synergistic gain between data resources and computing power resources, as well as the deviation penalty between computing power resources and the computing power resources required by model resources, can be emphasized to improve the overall application effect of the system. For example, when the market opens and trading volume surges, resource status data is obtained. Due to CPU processing queue backlog, i.e., insufficient computing power resources, the scheduling priority and adjustment parameters of each resource are obtained, indicating that computing power resources should be adjusted first. Resource status data is obtained again. Due to the surge in data volume, through analysis and calculation, it is determined that data resources should be adjusted first subsequently. After obtaining resource status data again, the adjustment direction and magnitude of model resources are determined. Finally, adjustments were made sequentially based on the direction and magnitude of each resource adjustment. The adjustment process was as follows: By activating high-energy-efficiency hardware, general computing tasks were offloaded to high-energy-efficiency hardware for processing. By increasing computing power, the energy efficiency of model business processing was reduced, resulting in the optimal improvement in system performance at this step. For scenarios where massive amounts of data cannot be fully and deeply cleaned, adaptive sampling was enabled. For small-amount, low-risk merchant transactions, only core features were extracted; for large-amount, abnormal transactions, all features were retained. This reduced data quality in exchange for lower model business processing efficiency and reduced computing power requirements. Finally, the model resources were switched to a cascaded model. The first-level model, using logistic regression, ran on high-energy-efficiency hardware handling 90% of the traffic. Only transactions marked as suspicious by the first-level model were routed to the second-level model for fine-tuning. The second-level model, a deep neural network, ran on a GPU. Through this complete scheduling process of the three resources in the intelligent system, the system reduced overall computing energy consumption while ensuring risk control coverage and avoiding latency defaults, maximizing the first function value to improve the overall application effect of the financial high-frequency trading risk control system.

[0125] The technical solution of this invention determines the first scheduling priority of each resource based on first data and a first function, and selects the highest-priority first resource for adjustment based on the first scheduling priority. This quantifies and distinguishes the contribution of different resources to the overall system performance, achieving scientific and orderly scheduling decisions. It also enables resource scheduling to prioritize the resource objects that have the most significant impact on system performance. Determining the second scheduling priority of the remaining resources based on the second data after the first resource adjustment allows for dynamic adaptation to the actual system operation scenario after resource status changes, avoiding poor adaptability and redundant adjustments caused by a fixed scheduling order. Selecting and adjusting the second resource based on the second scheduling priority continuously prioritizes the resource objects that have the most significant impact on system performance in a single iteration, further optimizing the improvement in system performance. Adjusting the remaining third resources based on the third data completes the closed-loop scheduling of the three resources, avoiding the problem of uncoordinated overall system configuration after adjusting a single resource, improving the matching degree and rationality of system resource configuration, and thus enhancing the overall application effect of the intelligent system.

[0126] Figure 3 This is a schematic diagram of a smart system resource scheduling device provided in an embodiment of the present invention. This embodiment is applicable to scheduling various resources of a smart system in various smart system application scenarios. The smart system resource scheduling device can be implemented in hardware and / or software, and can be configured in any electronic device with network communication capabilities. Figure 3 As shown, the intelligent system resource scheduling device provided in this embodiment of the invention may include the following:

[0127] The data acquisition module 310 is used to acquire the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at a first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing power resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in the running state.

[0128] The scheduling information generation module 320 is used to generate first resource scheduling information of the first intelligent system based on the first data and through a first function. The first function is a pre-constructed target optimization function, with a first variable, a second variable, and a third variable as independent variables. The first function calculates the total contribution of data resources, computing power resources, and model resources to the application effect of the intelligent system. The first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the computing power resources already in use in the intelligent system, and the third variable is the business processing quality of the model resources in operation in the intelligent system. The first resource scheduling information includes the scheduling priorities and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system. The adjustment parameters indicate the direction of adjustment for each resource configuration, and the scheduling priorities are positively correlated with the contribution of each resource configuration adjustment in the intelligent system to improving the application effect of the intelligent system.

[0129] The resource scheduling module 330 is used to schedule the data resources, computing resources and model resources in the first intelligent system according to the first resource scheduling information.

[0130] Based on the above embodiments, optionally, the first function includes a first function term and a second function term. The first function term is used to calculate the weighted sum of the contribution values ​​of data resources, computing power resources and model resources to the application effect of the intelligent system. The second function term is used to calculate the weighted sum of the contribution values ​​of multiple resource combinations to the application effect of the intelligent system. The resource combination is a combination of any two of the data resources, computing power resources and model resources.

[0131] Based on the above embodiments, optionally, the second function term includes a first basis function, a second basis function, and a third basis function. The first basis function is a logarithmic growth function used to calculate the contribution of the interaction between data resources and computing power resources to the application effect of the intelligent system. The second basis function is a saturated growth function used to calculate the contribution of the interaction between data resources and model resources to the application effect of the intelligent system. The third basis function is a deviation penalty function used to calculate the contribution of the matching degree between the first computing power and the second computing power to the application effect of the intelligent system. The first computing power is the computing power resource that has been enabled, and the second computing power is the computing power resource required by the model resource that is in operation.

[0132] Based on the above embodiments, optionally, the first resource scheduling information of the first intelligent system is generated by a first function according to the first data, including:

[0133] Based on the first data and the first function, determine the first scheduling priority of each resource in the first intelligent system;

[0134] Based on the first scheduling priority of each resource in the first intelligent system, the first resource is determined, and the first resource is the resource with the highest first scheduling priority in the first intelligent system.

[0135] Based on the first data and the first function, the value of the variable corresponding to the first resource is adjusted to obtain the second data. The second data includes the resource status data of the first intelligent system at a second time. The second time is the time when the value of the variable corresponding to the first resource is adjusted.

[0136] Based on the second data and the first function, determine the second scheduling priority of each remaining resource other than the first resource;

[0137] Based on the second scheduling priority, a second resource is determined, which is the resource with the highest second scheduling priority in the first intelligent system;

[0138] Based on the second data and the first function, the value of the variable corresponding to the second resource is adjusted to obtain the third data. The third data includes the resource status data of the first intelligent system at the third moment, and the third moment is the moment when the value of the variable corresponding to the second resource is adjusted.

[0139] Based on the third data and the first function, the variable value corresponding to the third resource is adjusted. The third resource is a resource in the first intelligent system other than the first resource and the second resource.

[0140] Based on the above embodiments, optionally, determining the first scheduling priority of each resource in the first intelligent system according to the first data and the first function includes:

[0141] Based on the first data and the first function, a first parameter, a second parameter, and a third parameter are determined. The first parameter is the partial derivative of the first function with respect to a first variable. The second parameter is the partial derivative of the first function with respect to a second variable. The third parameter is the gain value of the second model resource. The second model resource is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency indicated in the first data. The gain value is the difference between the first function value after predicting the model switch and the current first function value. The current first function value is calculated based on the first data.

[0142] Based on the first parameter, the second parameter, the third parameter, and the preset first weight combination, the first scheduling priority of each resource in the first intelligent system is determined. The first scheduling priority of data resources is the product of the absolute value of the first parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of computing power resources is the product of the absolute value of the second parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of model resources is the product of the third parameter and the corresponding weight in the preset first weight combination.

[0143] Based on the above embodiments, optionally, generating the first resource scheduling information of the first intelligent system through a first function based on the first data further includes:

[0144] Based on the first parameter, the second parameter, and the preset step size, a first adjustment parameter and a second adjustment parameter are determined. The first adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the data resource. The second adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the computing power resource. The first parameter is the partial derivative value of the first function with respect to the first variable, and the second parameter is the partial derivative value of the first function with respect to the second variable.

[0145] Based on the first data and the first function, a third adjustment parameter is determined. The third adjustment parameter is used to indicate the second model resource, which is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency in the first data. The gain value is the difference between the first function value after predicting the switching of model resources and the current first function value. The current first function value is calculated based on the first data.

[0146] Based on the above embodiments, optionally, after scheduling the data resources, computing power resources, and model resources in the first intelligent system according to the first resource scheduling information, the intelligent system resource scheduling device provided in this embodiment of the invention may further include:

[0147] The function value calculation module is used to calculate the first function value at the first moment based on the first data and the first function.

[0148] The data acquisition module 310 is also used to acquire fourth data, which includes the resource status data of the first intelligent system at a fourth moment, wherein the fourth moment is the moment after the scheduling of each resource in the first intelligent system is completed according to the first resource scheduling information;

[0149] The function value calculation module is used to calculate the first function value at the fourth time step based on the fourth data and the first function.

[0150] The gain value calculation module is used to calculate the gain value of this resource scheduling based on the first function value at the first time and the first function value at the fourth time.

[0151] The counting module is used to increment the cumulative number of invalid resource schedulings by one if the gain value of the current resource scheduling is not greater than a preset gain threshold.

[0152] The counting module is also used to reset the cumulative number of invalid resource scheduling to zero if the gain value of the current resource scheduling is greater than the preset gain threshold.

[0153] The resource scheduling module 330 is further configured to stop resource scheduling if the cumulative number of invalid resource scheduling is not less than a preset scheduling number threshold and the preset step size is less than a preset minimum step size, wherein the preset step size is used to indicate the degree of configuration adjustment for each resource scheduling.

[0154] The resource scheduling module 330 is also used to, if the cumulative number of invalid resource scheduling is less than the preset scheduling number threshold, reduce the preset step size by a preset ratio, and return to reacquire the resource status data of the first intelligent system for the next resource scheduling.

[0155] The technical solution of this invention, by acquiring the first data of the first intelligent system, can accurately grasp the current data resource quality, computing power efficiency, and business processing quality of the operating model, providing comprehensive and realistic status support for resource scheduling. Based on the first data and using a target optimization function with three types of resource indicators as independent variables to generate first resource scheduling information, it can quantify the total contribution of data, computing power, and model resources to the system application effect, making the determination of scheduling priorities and adjustment parameters more objective and accurate, ensuring a high degree of matching between scheduling strategies and resource contribution levels, thereby improving the pertinence and effectiveness of resource scheduling decisions and avoiding scheduling deviations caused by subjective judgment. Scheduling data resources, computing power resources, and model resources separately based on the first resource scheduling information can avoid resource allocation imbalances and ineffective scheduling, improve the efficiency of collaborative utilization of various resources, and thus enhance the overall operational stability and processing capacity of the intelligent system. Based on the above technical solutions, by collecting resource status in multiple dimensions and driving scheduling with multi-objective optimization functions, the problems of unreasonable resource allocation and unclear adjustment direction in the resource scheduling process of intelligent systems can be solved. It can realize the fine-grained collaborative scheduling of data resources, computing power resources and model resources, thereby improving the adaptive scheduling capability of intelligent systems and improving the overall energy efficiency and business processing quality of the system.

[0156] The intelligent system resource scheduling device provided in this embodiment of the invention can execute the intelligent system resource scheduling method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0157] The acquisition, storage, use, and processing of data in this application comply with relevant national laws and regulations and do not violate public order and good morals.

[0158] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0159] Figure 4 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0160] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0161] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0162] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods described above, such as intelligent system resource scheduling methods.

[0163] In some embodiments, the intelligent system resource scheduling method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the intelligent system resource scheduling method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the intelligent system resource scheduling method by any other suitable means (e.g., by means of firmware).

[0164] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific reference products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0165] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0166] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0167] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0168] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as data servers), or computing systems that include switching components (e.g., application servers), or computing systems that include front-end components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such back-end, switching, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0169] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0170] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0171] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent system resource scheduling method provided in any embodiment of this application.

[0172] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0173] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0174] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for intelligent system resource scheduling, the method comprising: The method includes: Acquire first data of the first intelligent system. The first data includes resource status data of the first intelligent system at a first moment. The resource status data includes the quality of data resources used by the first intelligent system, the energy efficiency of the enabled computing power resources, and the business processing quality of the first model resources. The first model resources are model resources in the first intelligent system that are in operation. Based on the first data, first resource scheduling information of the first intelligent system is generated through a first function; wherein, the first function is a pre-constructed target optimization function, the first function takes a first variable, a second variable, and a third variable as independent variables, the first function is used to calculate the total contribution value of data resources, computing power resources, and model resources to the application effect of the intelligent system, the first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the computing power resources that have been enabled in the intelligent system, and the third variable is the business processing quality of the model resources that are in operation in the intelligent system; the first resource scheduling information includes the scheduling priority and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system, the adjustment parameters are used to indicate the adjustment direction of each resource configuration, and the scheduling priority is positively correlated with the contribution of each resource configuration adjustment in the intelligent system to the improvement of the application effect of the intelligent system; Based on the first resource scheduling information, the data resources, computing resources, and model resources in the first intelligent system are scheduled respectively.

2. The method of claim 1, wherein, The first function includes a first function term and a second function term. The first function term is used to calculate the weighted sum of the contribution values ​​of data resources, computing power resources and model resources to the application effect of the intelligent system. The second function term is used to calculate the weighted sum of the contribution values ​​of multiple resource combinations to the application effect of the intelligent system. The resource combination is a combination of any two of the data resources, computing power resources and model resources.

3. The method of claim 2, wherein, The second function term includes a first basis function, a second basis function, and a third basis function. The first basis function is a logarithmic growth function, used to calculate the contribution of the interaction between data resources and computing power resources to the application effect of the intelligent system. The second basis function is a saturated growth function, used to calculate the contribution of the interaction between data resources and model resources to the application effect of the intelligent system. The third basis function is a bias penalty function, used to calculate the contribution of the matching degree between the first computing power and the second computing power to the application effect of the intelligent system. The first computing power is the computing power resource that has been enabled, and the second computing power is the computing power resource required by the model resource that is in operation.

4. The method of claim 1, wherein, Based on the first data, the first resource scheduling information of the first intelligent system is generated through a first function, including: Based on the first data and the first function, determine the first scheduling priority of each resource in the first intelligent system; Based on the first scheduling priority of each resource in the first intelligent system, the first resource is determined, and the first resource is the resource with the highest first scheduling priority in the first intelligent system. Based on the first data and the first function, the value of the variable corresponding to the first resource is adjusted to obtain the second data. The second data includes the resource status data of the first intelligent system at a second time. The second time is the time when the value of the variable corresponding to the first resource is adjusted. Based on the second data and the first function, determine the second scheduling priority of each remaining resource other than the first resource; Based on the second scheduling priority, a second resource is determined, which is the resource with the highest second scheduling priority in the first intelligent system; Based on the second data and the first function, the value of the variable corresponding to the second resource is adjusted to obtain the third data. The third data includes the resource status data of the first intelligent system at the third moment, and the third moment is the moment when the value of the variable corresponding to the second resource is adjusted. Based on the third data and the first function, the variable value corresponding to the third resource is adjusted. The third resource is a resource in the first intelligent system other than the first resource and the second resource.

5. The method of claim 4, wherein, Based on the first data and the first function, the first scheduling priority of each resource in the first intelligent system is determined, including: Based on the first data and the first function, a first parameter, a second parameter, and a third parameter are determined. The first parameter is the partial derivative of the first function with respect to a first variable. The second parameter is the partial derivative of the first function with respect to a second variable. The third parameter is the gain value of the second model resource. The second model resource is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency indicated in the first data. The gain value is the difference between the first function value after predicting the model switch and the current first function value. The current first function value is calculated based on the first data. Based on the first parameter, the second parameter, the third parameter, and the preset first weight combination, the first scheduling priority of each resource in the first intelligent system is determined. The first scheduling priority of data resources is the product of the absolute value of the first parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of computing power resources is the product of the absolute value of the second parameter and the corresponding weight in the preset first weight combination. The first scheduling priority of model resources is the product of the third parameter and the corresponding weight in the preset first weight combination.

6. The method according to claim 4 or 5, characterized in that, Based on the first data, generating the first resource scheduling information of the first intelligent system through a first function further includes: Based on the first parameter, the second parameter, and the preset step size, a first adjustment parameter and a second adjustment parameter are determined. The first adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the data resource. The second adjustment parameter is used to indicate the adjustment direction and degree of the variable value corresponding to the computing power resource. The first parameter is the partial derivative value of the first function with respect to the first variable, and the second parameter is the partial derivative value of the first function with respect to the second variable. Based on the first data and the first function, a third adjustment parameter is determined. The third adjustment parameter is used to indicate the second model resource, which is the model resource with the largest gain value among the third model resources. The third model resource is the model resource that can be supported by the data resource quality and computing power efficiency in the first data. The gain value is the difference between the first function value after predicting the switching of model resources and the current first function value. The current first function value is calculated based on the first data.

7. The method of claim 1, wherein, After scheduling the data resources, computing resources, and model resources in the first intelligent system according to the first resource scheduling information, the method further includes: Based on the first data, the first function value at the first moment is calculated using the first function; Obtain fourth data, which includes the resource status data of the first intelligent system at the fourth moment, where the fourth moment is the moment after the scheduling of each resource in the first intelligent system is completed according to the first resource scheduling information; Based on the fourth data, the value of the first function at the fourth time point is calculated using the first function. Calculate the gain value of this resource scheduling based on the first function value at the first time point and the first function value at the fourth time point; If the gain value of this resource scheduling is not greater than the preset gain threshold, the cumulative number of invalid resource scheduling is incremented by one; If the gain value of the current resource scheduling is greater than the preset gain threshold, the cumulative number of invalid resource scheduling attempts is reset to zero. If the cumulative number of invalid resource schedulings is not less than a preset scheduling number threshold, and the preset step size is less than a preset minimum step size, resource scheduling is stopped. The preset step size is used to indicate the degree of configuration adjustment for each resource scheduling. If the cumulative number of invalid resource scheduling attempts is less than the preset scheduling attempt threshold, the preset step size is decayed according to a preset ratio, and the system returns to reacquire the resource status data of the first intelligent system for the next resource scheduling attempt.

8. An intelligent system resource scheduling apparatus, characterized by, The device includes: The data acquisition module is used to acquire the first data of the first intelligent system. The first data includes the resource status data of the first intelligent system at a first moment. The resource status data includes the quality of the data resources used by the first intelligent system, the energy efficiency of the enabled computing power resources, and the business processing quality of the first model resources. The first model resources are the model resources in the first intelligent system that are in the running state. The scheduling information generation module is used to generate first resource scheduling information of the first intelligent system based on the first data and through a first function. The first function is a pre-constructed target optimization function, with a first variable, a second variable, and a third variable as independent variables. The first function calculates the total contribution of data resources, computing power resources, and model resources to the application effect of the intelligent system. The first variable reflects the quality of data resources in the intelligent system, the second variable is the energy efficiency generated by the enabled computing power resources in the intelligent system, and the third variable is the business processing quality of the model resources in operation in the intelligent system. The first resource scheduling information includes the scheduling priorities and adjustment parameters of data resources, computing power resources, and model resources in the first intelligent system. The adjustment parameters indicate the adjustment direction of each resource configuration, and the scheduling priority is positively correlated with the contribution of each resource configuration adjustment in the intelligent system to improving the application effect of the intelligent system. The resource scheduling module is used to schedule the data resources, computing resources and model resources in the first intelligent system according to the first resource scheduling information.

9. An electronic device, comprising: The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the intelligent system resource scheduling method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the intelligent system resource scheduling method according to any one of claims 1-7.

Images