Resource management system based on artificial intelligence platform

Abstract

The invention is suitable for the technical field of resource management, and provides a resource management system based on an artificial intelligence platform, which comprises a resource demand end, a resource scheduling center and a resource supply end, the resource demand end comprises a resource demand identification module used for identifying the resource type of the demand; the resource scheduling center comprises a resource demand receiving module, a resource supply feedback module, a resource state analysis module and a resource scheduling module; the resource demand receiving module is used for receiving demand information sent by the resource demand end; the resource supply feedback module is used for receiving supply information fed back by the demand supply end; the resource state analysis module is used for analyzing the state change of resource demand and supply; the resource scheduling module is used for performing resource scheduling according to the analyzed resource state change; the resource supply end comprises a resource checking module which is used for counting available resource information. According to the invention, dynamic resource scheduling can be carried out based on state change, and the resource utilization rate is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of resource management, and in particular to a resource management system based on an artificial intelligence platform. Background Art

[0002] With the widespread application of technologies such as big data, artificial intelligence and machine learning, the demand for high-performance computing continues to grow. Computing resources such as server clusters, graphics processors and application-specific integrated circuits have become the main resources, and data centers, as the main carrier of computing resources, are driving technological innovation. Whether it is computing power and storage resources in IT resources, or human and material resources in other fields, if there is a lack of effective management, it is easy to waste.

[0003] In the prior art, resource allocation and scheduling involved in resource management allocate resources according to priority, and the plan can be dynamically adjusted according to the actual usage and demand of resources. However, this allocation method is relatively dependent on resource monitoring tools, which require real-time collection of usage information of each resource. In general, there is a certain time difference between the monitoring tool transmitting the monitoring data to the resource management system and processing it. When the resource status changes frequently, the management system will make inaccurate resource allocation decisions based on outdated data, and even cause problems such as low resource utilization or resource waste.

[0004] In view of this, a resource management system based on an artificial intelligence platform is needed. Summary of the invention

[0005] The embodiment of the present application provides a resource management system based on an artificial intelligence platform, which is used to solve the problem of low resource utilization or resource waste.

[0006] A first aspect of an embodiment of the present application provides a resource management system based on an artificial intelligence platform, including:

[0007] A resource demand side, a resource scheduling center and a resource supply side, wherein the resource scheduling center is respectively connected to the resource demand side and the resource supply side in communication;

[0008] The resource demand end includes a resource demand identification module, and the demand identification module is used to identify the resource type required, and the resource type includes computing power resources, storage resources and network resources;

[0009] The resource scheduling center includes a resource demand receiving module, a resource supply feedback module, a resource status analysis module and a resource scheduling module. The resource demand receiving module is used to receive demand information sent by the resource demand side; the resource supply feedback module is used to receive supply information fed back by the demand supply side; the resource status analysis module is used to analyze the state changes of resource demand and supply; the resource scheduling module is used to perform resource scheduling according to the analyzed resource state changes;

[0010] The resource supply end includes a resource inventory module, and the resource inventory module is used to count the supplyable resource information.

[0011] Furthermore, the demand identification module is used to identify the resource type of the demand, and the resource type includes computing resources, storage resources and network resources, including:

[0012] Determine resource type related words based on the identified demand information, classify the resource type related words according to the resource type, and construct a keyword dictionary of each type;

[0013] Extracting keywords after performing grammatical analysis and part-of-speech tagging on the text in the keyword dictionary;

[0014] The required resource type is determined by classifying and matching the extracted keywords.

[0015] Furthermore, the resource status analysis module is used to analyze the status changes of resource demand and supply, including:

[0016] Determining whether the demand information sent by the resource demand end is marked information;

[0017] If so, the change in resource quantity, the change in resource quality and the change in resource time periodicity corresponding to the demand information are analyzed by using a gradient boosting regression tree algorithm;

[0018] If not, the target demand information is determined by correlation analysis, and the corresponding resource quantity change, resource quality change and resource time periodic change are determined according to the target demand information.

[0019] Furthermore, the analyzing the resource quantity change, resource quality change and resource time periodic change corresponding to the demand information by using the gradient boosting regression tree algorithm includes:

[0020] Determine resource quantity data, resource quality data and resource time data according to the resource type in the demand information, and divide the data into a training set and a test set;

[0021] Based on each resource type, the resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics are calculated respectively;

[0022] The resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics are used as training samples and input into the constructed GBRT model for iterative training;

[0023] The trained GBRT model is evaluated using the test set to obtain the predicted resource quantity changes, predicted resource quality changes, and predicted resource time periodic changes corresponding to each resource type.

[0024] Furthermore, the trained GBRT model is evaluated using the test set to obtain the predicted resource quantity change, predicted resource quality change and predicted resource time periodic change corresponding to each resource type, including:

[0025]

[0026] Where: m is the set number of iterations, is the predicted value of the resource status change of the i-th element, v is the learning rate, f m (X i ) is a decision tree representing quantity changes, f m (Y i ) is a decision tree representing quality change, f m (Z i ) is a decision tree representing time periodic changes.

[0027] Furthermore, if not, the target demand information is determined by correlation analysis, and the corresponding resource quantity change, resource quality change and resource time periodic change are determined according to the target demand information, including:

[0028] When the demand information sent by the resource demand end is unlabeled information, the demand information is matched with the labeled demand information by similarity through a correlation analysis method, and the demand information that meets the similarity condition is selected as the target demand information;

[0029] The corresponding resource quantity change, resource quality change and resource time periodic change are calculated according to the target demand information.

[0030] Furthermore, the resource scheduling module is used to perform resource scheduling according to the analyzed resource status changes, including:

[0031] When the identified demand resource types include at least one type, a resource demand assessment system is constructed, wherein the criterion layer of the resource demand assessment system includes computing power resource types, storage resource types, and network resource types, and the sub-criterion layers of the computing power resource types, storage resource types, and network resource types respectively include predicted resource quantity changes, predicted resource quality changes, and predicted resource time periodic changes;

[0032] The subjective weight of each indicator element in the resource demand assessment system is calculated using the hierarchical analysis method, and the objective weight of each indicator element in the resource demand assessment system is calculated using the entropy weight method;

[0033] Calculate the comprehensive demand weight of each resource type based on the subjective weight and the objective weight;

[0034] The target demand supply side is screened according to the comprehensive demand weights for resource scheduling.

[0035] Furthermore, the indicator layer for predicting resource quantity changes in the computing power resource type includes the number of computing cores and accelerators, and memory capacity; the indicator layer for predicting resource quality changes in the computing power resource type includes CPU and GPU performance, and computing architecture; the indicator layer for predicting resource time periodic changes in the computing power resource type includes task response and completion time, and changes in the proportion of resource idle and busy time; the indicator layer for predicting resource quality changes in the storage resource type includes storage capacity, the number of storage devices, and the number and size of storage partitions; the indicator layer for predicting resource quality changes in the storage resource type includes read and write speed, data reliability, and storage medium life; the indicator layer for predicting resource time periodic changes in the storage resource type includes data storage cycle, data backup and recovery time, storage resource space, and usage time; the indicator layer for predicting resource quantity changes in the network resource type includes network bandwidth, the number of network ports and IP addresses; the indicator layer for predicting resource quality changes in the network resource type includes network delay, packet loss rate, and network jitter; the indicator layer for predicting resource time periodic changes in the network resource type includes network peak and trough time, data transmission time, network resource space, and busy time ratio.

[0036] Furthermore, the comprehensive demand weight of each resource type is calculated based on the subjective weight and the objective weight:

[0037]

[0038] Where: w A and are the subjective weight vector and objective weight vector of the criterion layer, and are the subjective weight vector and objective weight vector of the sub-criteria layer under the k-th resource type, and are the subjective weight vector and objective weight vector of the indicator layer under the lth sub-criteria layer under the kth resource type, respectively, and α is the combination coefficient.

[0039] Furthermore, the step of screening the target demand supply end according to the comprehensive demand weight to perform resource scheduling includes:

[0040] Determine input data of the hash algorithm based on the supply information fed back by the resource supply end, and construct a data structure;

[0041] Determine a hash function according to the weights of the computing resource type, storage resource type, and network resource type in the resource evaluation system, and calculate the corresponding hash value;

[0042] The target resource supplier is screened according to the hash value.

[0043] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0044] 1. The resource status analysis module in the resource management system of the present invention can obtain in real time various changes of resources within a certain period of time in the future by analyzing the changes in the status of resource demand and supply, thereby solving problems such as untimely resource allocation, increased risk of business interruption, and difficulty in cost control caused by frequent changes in status. It can provide more accurate data basis for resource matching and provide strong guarantee for enterprises to achieve efficient resource management and stable business operations.

[0045] 2. The resource scheduling module in the resource management system of the present invention analyzes the variables and features involved in scheduling according to the resource status changes of each resource type, and can perform dynamic resource scheduling based on status changes, effectively improving the real-time and accuracy of resource scheduling.

[0046] Other advantages, objectives, and features of the present invention will be set forth in part in the following description, and in part will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a structural block diagram of an embodiment of a resource management system of an artificial intelligence platform in the present invention;

[0048] Figure 2 This is a schematic diagram of an embodiment of a demand identification module in the present invention;

[0049] Figure 3 It is a schematic diagram of an embodiment flow of a resource status analysis module in the present invention;

[0050] Figure 4 This is a schematic diagram of an implementation flow of a resource scheduling module in the present invention;

[0051] Figure 5 It is a structural diagram of the resource demand evaluation system of the resource scheduling module in the present invention. DETAILED DESCRIPTION

[0052] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0053] Embodiment 1

[0054] The following will introduce the resource management system based on the artificial intelligence platform in the present invention from the perspective of information interaction between the resource demand side, the resource scheduling center and the resource supply side. Figure 1 , a resource management system based on an artificial intelligence platform includes:

[0055] Resource demand side 1, resource scheduling center 2 and resource supply side 3, resource scheduling center 2 is respectively connected to resource demand side 1 and resource supply side 3 in communication;

[0056] The resource demand side 1 includes a resource demand identification module 101, which is used to identify the resource type of the demand, and the resource type includes computing resources, storage resources and network resources; the resource scheduling center 2 includes a resource demand receiving module 201, a resource supply feedback module 202, a resource status analysis module 203 and a resource scheduling module 204, the resource demand receiving module 201 is used to receive the demand assessment information sent by the resource demand side; the resource supply feedback module 202 is used to receive the supply information fed back by the demand supply side; the resource status analysis module 203 is used to analyze the state changes of resource demand and supply; the resource scheduling module 204 is used to perform resource scheduling according to the analyzed resource state changes; the resource supply side 3 includes a resource inventory module 301, and the resource inventory module 301 is used to count the supplyable resource information.

[0057] Specifically, after receiving the demand information sent by each business platform, the resource demand end 1 mainly identifies it from the perspective of operation behavior, application system and tool, and performance bottleneck and problem feedback. After receiving the demand information sent by the resource demand end 1, the resource demand receiving module 201 of the resource scheduling center 2 publishes the demand information to each resource supply end 3, so that the resource supply end 3 performs statistical inventory based on the demand information and then sends the statistical resource information to the resource scheduling center 2. The resource status analysis module 203 analyzes the changes in the status of resource demand and supply based on the feedback supply information, and the resource scheduling module 204 finally performs resource scheduling.

[0058] The above embodiment implements real-time information management of resources through the resource scheduling center 2 and the resource demand side 1 and the resource supply side 3. Through information data interaction, accurate matching and efficient allocation of resources can be achieved, resource utilization efficiency can be improved, and stable operation and sustainable development of the business can be guaranteed.

[0059] Embodiment 2

[0060] See also Figure 2 The demand identification module is used to identify the resource types required, including computing resources, storage resources, and network resources, including the following steps:

[0061] S21. Determine resource type-related vocabulary based on the identified demand information, classify resource type-related vocabulary according to resource type, and construct a keyword dictionary for each type;

[0062] Specifically, for example, for computing resource types, words such as "calculation", "computation", "processing", "simulation", "rendering", "CPU", "GPU", "parallel computing", and "high-performance computing" are collected. These words represent the operations and hardware that may be involved in computing resource requirements. For example, storage resource types include words such as "storage", "save", "backup", "capacity", "hard disk", "storage array", "data warehouse", and "read and write speed". For example, network resource types, words such as "transmission", "communication", "bandwidth", "delay", "packet loss", "network topology", "access", "Wi-Fi", and "optical fiber" are selected. These words reflect the main characteristics of network resources.

[0063] S22. Extract keywords after performing grammatical analysis and part-of-speech tagging on the text in the keyword dictionary;

[0064] Classify the collected words according to resource types to form a computing resource keyword dictionary, a storage resource keyword dictionary, and a network resource keyword dictionary. The words in each dictionary can be weighted according to their importance, frequency of occurrence, and other factors. Then perform text cleaning, lexical analysis, and part-of-speech tagging. Calculate the frequency of occurrence of each word in the demand text. A simple counting method can be used to count the number of times each word appears. For example, in the cleaned text, "data calculation" appears once and "storage data" appears once. If there is a text collection consisting of multiple demand texts, TF-IDF can be used to calculate the importance of each word. Select the top-ranked words as keywords based on word frequency or TF-IDF value. A threshold can be set, for example, only select words with a word frequency greater than 1 or a TF-IDF value greater than a set value. At the same time, consider the completeness of the vocabulary and avoid extracting overly fragmented words. For example, "data calculation" makes more sense as a complete concept rather than separate "data" and "calculation".

[0065] S23. Determine the required resource type after classification and matching based on the extracted keywords.

[0066] Match the extracted keywords with the previously constructed keyword dictionary. For each keyword, find the resource type in which it appears. For example, "data computing" finds a match in the computing power resource keyword dictionary, and "storage data" finds a match in the storage resource keyword dictionary. Based on the matched keywords and their weights, calculate the score of each resource type. Compare the scores of each resource type, and the resource type with the highest score is the type to which the requirement belongs.

[0067] Embodiment 3

[0068] See also Figure 3 ,The resource status analysis module is used to analyze the status changes of resource demand and supply, including the following steps:

[0069] S31. Determine whether the demand information sent by the resource demand side is marked information;

[0070] Specifically, the purpose of determining whether the demand information sent by the resource demand side is marked information is to find out whether similar demands have been initiated before the demand information, and to determine whether to modify or change to a new business demand based on the original business.

[0071] S32. When the demand information sent by the resource demand side is marked information, the resource quantity data, resource quality data and resource time data are determined according to the resource type in the demand information, and the data is divided into a training set and a test set;

[0072] When the demand information sent by the resource demand side is labeled information, collect the labeled historical demand information and the corresponding resource data, including the number of resources at different time points (such as the number of servers in computing resources, the capacity of storage resources, the number of connections of network resources, etc.), quality indicators (such as the CPU performance of computing resources, the read and write speed of storage resources, the bandwidth and latency of network resources, etc.) and timestamps, etc. Divide the data into training sets and test sets, usually according to a certain ratio.

[0073] S33. Calculate the resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics based on each resource type;

[0074] Specifically, calculate the resource quantity change characteristics: for example, for each resource type, calculate the resource quantity change rate at adjacent time points, such as (current quantity - quantity at the previous time point) / quantity at the previous time point. This change rate can be used as a feature to reflect the dynamic change trend of the resource quantity.

[0075] Computing resource quality change characteristics: For example, for computing resources, the changes in performance indicators such as CPU main frequency and number of cores can be extracted; for storage resources, the changes in indicators such as read and write speed and storage capacity utilization can be extracted; for network resources, the changes in indicators such as bandwidth, latency, and packet loss rate can be extracted. For example, for storage resources, the changes in mean time between failures (MTBF) can be calculated; for network resources, the changes in network availability can be counted.

[0076] Calculate the time-periodic change characteristics of resources: Encode the timestamp, for example, extract information such as month, quarter, whether it is a holiday, etc.; for resource requirements with obvious periodicity, etc.

[0077] S34. The resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics are used as training samples and input into the constructed GBRT model for iterative training;

[0078] S35. Evaluate the trained GBRT model using the test set to obtain the predicted resource quantity change, predicted resource quality change, and predicted resource time periodicity change corresponding to each resource type;

[0079] 1. Initialize a very simple model, usually a constant value, such as the initial prediction value is the mean of all sample target values.

[0080] 2. Perform iterative training, for m=1 to M, M is the set number of iterations.

[0081] 3. Calculate the negative gradient of the loss function as the residual, and assume that the loss function is Usually the squared loss function can be used Then the residual is:

[0082]

[0083] 4. Use decision trees to fit residuals: Construct three decision trees to fit residuals:

[0084]

[0085] Where: m is the set number of iterations, is the predicted value of the resource status change of the i-th element, v is the learning rate, f m (X i ) is a decision tree representing quantity changes, f m (Y i ) is a decision tree representing quality change, f m (Z i ) is a decision tree representing time periodic changes.

[0086] 6. Evaluate model performance using the test set The evaluation metric used here is mean square error (MSE).

[0087] S36. When the demand information sent by the resource demand side is unlabeled information, the target demand information is determined by correlation analysis, and the corresponding resource quantity change, resource quality change and resource time periodic change are determined according to the target demand information.

[0088] Step S36 includes the following:

[0089] When the demand information sent by the resource demand side is unlabeled information, the demand information is matched with the labeled demand information through correlation analysis, and the demand information that meets the similarity conditions is selected as the target demand information; the corresponding resource quantity changes, resource quality changes and resource time periodic changes are calculated based on the target demand information.

[0090] Specifically, when the demand information sent by the resource demand side is unlabeled information, the correlation analysis method is used to find labeled demand information with a high degree of similarity, and the corresponding resource quantity changes, resource quality changes, and resource time periodic changes are estimated through the labeled demand information. Here, the demand information with high similarity is found through correlation. For example, when the demand business is a new business, the scheduling decision can be dynamically adjusted according to the estimated resource status changes.

[0091] Embodiment 4

[0092] See also Figure 4 ,The resource scheduling module is used to schedule resources according to the analyzed resource status changes, including the following steps:

[0093] S41. When the identified demand resource types include at least one type, a resource demand assessment system is constructed, wherein the criterion layer of the resource demand assessment system includes computing resource types, storage resource types, and network resource types, and the sub-criterion layers of computing resource types, storage resource types, and network resource types respectively include predicted resource quantity changes, predicted resource quality changes, and predicted resource time periodic changes;

[0094] It should be noted that this embodiment is based on the evaluation system being constructed only when the identified demand resource types include at least one type. When the identified demand resource type is one, the appropriate resource supplier can be matched directly by the quantity, quality and time periodic changes of the demand of this type. When there are two or three types, the supplier needs to be assigned after the weights are evaluated.

[0095] See also Figure 5 ,Here taking the three identified resource types as an example, the ,criterion layer of the resource demand assessment system includes computing ,resource type, storage resource type, and network resource type, and the sub-criteria layers of computing ,resource type, storage resource type, and network resource type include the predicted ,change in resource quantity, the predicted change in resource quality, and the predicted ,time periodicity of resources, respectively.

[0096] The indicator layer for predicting resource quantity changes in computing power resource types includes the number of computing cores and accelerators, and memory capacity; the indicator layer for predicting resource quality changes in computing power resource types includes CPU and GPU performance and computing architecture; the indicator layer for predicting resource time periodic changes in computing power resource types includes task response and completion time, and changes in the proportion of resource idle and busy time; the indicator layer for predicting resource quality changes in storage resource types includes storage capacity, the number of storage devices, and the number and size of storage partitions; the indicator layer for predicting resource quality changes in storage resource types includes read and write speed, data reliability, and storage medium life; the indicator layer for predicting resource time periodic changes in storage resource types includes data storage cycle, data backup and recovery time, storage resource space, and usage time; the indicator layer for predicting resource quantity changes in network resource types includes network bandwidth, the number of network ports and IP addresses; the indicator layer for predicting resource quality changes in network resource types includes network delay, packet loss rate, and network jitter; the indicator layer for predicting resource time periodic changes in network resource types includes network peak and trough time, data transmission time, network resource space, and busy time ratio.

[0097] S42. Calculate the subjective weight of each indicator element in the resource demand assessment system using the analytic hierarchy process, and calculate the objective weight of each indicator element in the resource demand assessment system using the entropy weight method;

[0098] 1. Determine subjective weights using the analytic hierarchy process (AHP):

[0099] 1. Construct a judgment matrix:

[0100] Target layer-criterion layer judgment matrix (A): Let the judgment matrix A = (a ij ), where i and j represent computing resource type, storage resource type, and network resource type respectively. The relative importance of each resource type for demand assessment is determined by expert scoring, for example, 12 Indicates the importance of computing resource type relative to storage resource type, with a value range of 1-9.

[0101] Criteria layer-subcriteria layer judgment matrix (B k ): Based on the computing resource type, storage resource type, and network resource type, judgment matrices B1, B2, and B3 are constructed respectively. Taking the computing resource type as an example, Among them, i and j represent the change in the quantity of predicted resources, the change in the quality of predicted resources, and the change in the time period of predicted resources, respectively. The relative importance of each sub-criteria is also determined by expert scoring.

[0102] Sub-criteria layer-indicator layer judgment matrix (C kl ): Construct a judgment matrix based on the indicator layer under each sub-criteria layer. For example, under the sub-criteria of predicting resource quantity changes in the computing power resource type, construct a judgment matrix Where i and j represent the number of computing cores and accelerators, and memory capacity, respectively.

[0103] 2. Calculate the weight vector:

[0104] The target layer-criterion layer weight vector (w A ): Calculate the maximum eigenvalue λ of the judgment matrix A max and the corresponding eigenvector w A , and normalize it. Calculate the consistency index CI = (A max ―n) / (n―1), find the corresponding average random consistency index RI, and calculate the consistency ratio CR=CI / RI. If CR<0.1, then the normalized w A is the subjective weight vector of the criterion layer; otherwise, the judgment matrix A is readjusted.

[0105] Criteria layer-subcriteria layer weight vector For each judgment matrix B k , calculate the maximum eigenvalue and the corresponding eigenvector, perform normalization to obtain the subjective weight vector of each sub-criteria layer relative to the criterion layer, and perform consistency check in the same way.

[0106] Sub-criteria layer-index layer weight vector The indicator layer judgment matrix under each sub-criteria layer is calculated to obtain the subjective weight vector of the indicator layer relative to the sub-criteria layer.

[0107] 2. Determine the objective weight using the entropy weight method:

[0108] 1. Calculate the entropy value of each indicator based on the standardized data:

[0109] Let the entropy value of the jth indicator be e j , Where k = 1 / ln(m), m is the sample size.

[0110] 2. Calculate objective weights:

[0111] Calculate the objective weight of the jth indicator Obtain the objective weight vector of each indicator layer.

[0112] S43. Calculate the comprehensive demand weight of each resource type based on the subjective weight and the objective weight;

[0113] Specifically, appropriate combination coefficients α and (1-α) are selected to combine subjective weights and objective weights. The value of the combination coefficient can be determined based on the degree of trust in subjective experience and objective data.

[0114]

[0115] Where: w A and are the subjective weight vector and objective weight vector of the criterion layer, and are the subjective weight vector and objective weight vector of the sub-criteria layer under the k-th resource type, and They are respectively the subjective weight vector and objective weight vector of the indicator layer under the lth sub-criteria layer under the kth resource type.

[0116] S44. Filter the target demand supply side according to the comprehensive demand weight and perform resource scheduling.

[0117] Step S44 includes the following:

[0118] 1. Determine the input data of the hash algorithm based on the supply information fed back by the resource supply side and build the data structure;

[0119] Specifically, based on the key information of each resource supply side, this information will be used as input data for the hash algorithm. Including but not limited to the resource supply capacity of the supply side: such as the number of computing cores and memory capacity in computing resources; storage capacity and read and write speed in storage resources; bandwidth and number of ports in network resources; supply cost: such as unit computing power cost, storage cost per GB, and rental price per Mbps of network bandwidth; supply stability indicators: such as the number of failures in the past period of time, service availability, etc.

[0120] Construct the information of each resource provider into a data structure, for example, using a dictionary or object to store it. Taking a dictionary as an example, for a resource provider, the data structure may be as follows:

[0121] {Supply side name: Supplier A, computing power information: {Number of computing cores: 100, memory capacity: 512, computing power cost: 0.1, computing power stability: 0.95}, storage information: {storage capacity: 1000, read and write speed: 100, storage cost: 0.01, storage stability: 0.98}, network information: {bandwidth: 1000, number of ports: 10, network cost: 0.001, network stability: 0.97}}

[0122] 2. Determine the hash function based on the weights of computing resource types, storage resource types, and network resource types in the resource evaluation system, and calculate the corresponding hash value;

[0123] Determine the hash function based on the various indicator elements in the demand assessment system. For example, assuming that the computing power resource weight is 0.4, the storage resource weight is 0.3, and the network resource weight is 0.3. The hash function can be:

[0124] hash_value = (0.4*(computing power comprehensive evaluation value) + 0.3*(storage comprehensive evaluation value) + 0.3*(network comprehensive evaluation value))

[0125] Among them, the comprehensive evaluation value of computing power can be a comprehensive value calculated based on factors such as the number of computing cores, memory capacity, computing power cost, and computing power stability; the same applies to the comprehensive evaluation value of storage and the comprehensive evaluation value of the network. For example, the comprehensive evaluation value of computing power can be calculated by simple weighting: comprehensive evaluation value of computing power = 0.3*(number of computing cores / maximum number of cores)+0.3*(memory capacity / maximum memory capacity)-0.2*(computing power cost / maximum cost)+0.2*(computing power stability).

[0126] For each resource supplier, a hash value is calculated according to a determined hash function. For example, for the supplier A mentioned above, the computing power comprehensive evaluation value, storage comprehensive evaluation value and network comprehensive evaluation value are calculated according to the above calculation method, and then substituted into the hash function to calculate the hash value.

[0127] 3. Filter the target resource supplier based on the hash value.

[0128] Finally, the screening threshold is determined according to the demand. This threshold can be pre-set by analyzing historical data or according to business requirements. For example, the supplier with a hash value greater than is set as the supplier that meets the basic requirements. The calculated hash value of each resource supplier is compared with the set screening threshold. If the hash value is greater than or equal to the threshold, the supplier is screened out as a supplier that preliminarily meets the requirements. For example, after calculation, the hash value of supplier A is 0.75, which is greater than the set threshold of 0.7, then supplier A is screened out as a potentially suitable resource supplier, that is, the target resource supplier.

[0129] It is understandable that those skilled in the art can, under the guidance of the above embodiments, combine various implementation methods in the above embodiments to obtain technical solutions of multiple implementation methods.

[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A resource management system based on an artificial intelligence platform, characterized in that: include: A resource demand side, a resource scheduling center and a resource supply side, wherein the resource scheduling center is respectively connected to the resource demand side and the resource supply side in communication; The resource demand end includes a resource demand identification module, and the demand identification module is used to identify the resource type required, and the resource type includes computing power resources, storage resources and network resources; The resource scheduling center includes a resource demand receiving module, a resource supply feedback module, a resource status analysis module and a resource scheduling module. The resource demand receiving module is used to receive demand information sent by the resource demand side; the resource supply feedback module is used to receive supply information fed back by the demand supply side; the resource status analysis module is used to analyze the state changes of resource demand and supply; the resource scheduling module is used to perform resource scheduling according to the analyzed resource state changes; The resource supply end includes a resource inventory module, and the resource inventory module is used to count the supplyable resource information.

2. The resource management system based on the artificial intelligence platform according to claim 1 is characterized in that: The demand identification module is used to identify the resource type required, and the resource type includes computing resources, storage resources and network resources, including: Determine resource type related words based on the identified demand information, classify the resource type related words according to the resource type, and construct a keyword dictionary of each type; Extracting keywords after performing grammatical analysis and part-of-speech tagging on the text in the keyword dictionary; The required resource type is determined by classifying and matching the extracted keywords.

3. The resource management system based on the artificial intelligence platform according to claim 1 is characterized in that: The resource status analysis module is used to analyze the status changes of resource demand and supply, including: Determining whether the demand information sent by the resource demand end is marked information; If so, the change in resource quantity, the change in resource quality and the change in resource time periodicity corresponding to the demand information are analyzed by using a gradient boosting regression tree algorithm; If not, the target demand information is determined by correlation analysis, and the corresponding resource quantity change, resource quality change and resource time periodic change are determined according to the target demand information.

4. The resource management system based on the artificial intelligence platform according to claim 3 is characterized in that: The analyzing the resource quantity change, resource quality change and resource time periodic change corresponding to the demand information by using the gradient boosting regression tree algorithm includes: Determine resource quantity data, resource quality data and resource time data according to the resource type in the demand information, and divide the data into a training set and a test set; Based on each resource type, the resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics are calculated respectively; The resource quantity change characteristics, resource quality change characteristics and resource time periodic change characteristics are used as training samples and input into the constructed GBRT model for iterative training; The trained GBRT model is evaluated using the test set to obtain the predicted resource quantity changes, predicted resource quality changes, and predicted resource time periodic changes corresponding to each resource type.

5. The resource management system based on the artificial intelligence platform according to claim 4 is characterized in that: The trained GBRT model is evaluated using the test set to obtain the predicted resource quantity change, predicted resource quality change and predicted resource time periodic change corresponding to each resource type, including: Where: m is the set number of iterations, is the predicted value of the resource status change of the i-th element, v is the learning rate, and f m (X i ) is a decision tree representing quantity changes, f m (Y i ) is a decision tree representing quality change, f m (Z i ) is a decision tree representing time periodic changes.

6. The resource management system based on the artificial intelligence platform according to claim 4 is characterized in that: If not, the target demand information is determined by correlation analysis, and the corresponding resource quantity change, resource quality change and resource time periodic change are determined according to the target demand information, including: When the demand information sent by the resource demand end is unlabeled information, the demand information is matched with the labeled demand information by similarity through a correlation analysis method, and the demand information that meets the similarity condition is selected as the target demand information; The corresponding resource quantity change, resource quality change and resource time periodic change are calculated according to the target demand information.

7. The resource management system based on an artificial intelligence platform according to any one of claims 1 to 6, characterized in that: The resource scheduling module is used to perform resource scheduling according to the analyzed resource status changes, including: When the identified demand resource types include at least one type, a resource demand assessment system is constructed, wherein the criterion layer of the resource demand assessment system includes computing power resource types, storage resource types, and network resource types, and the sub-criterion layers of the computing power resource types, storage resource types, and network resource types respectively include predicted resource quantity changes, predicted resource quality changes, and predicted resource time periodic changes; The subjective weight of each indicator element in the resource demand assessment system is calculated using the hierarchical analysis method, and the objective weight of each indicator element in the resource demand assessment system is calculated using the entropy weight method; Calculate the comprehensive demand weight of each resource type based on the subjective weight and the objective weight; The target demand supply side is screened according to the comprehensive demand weights for resource scheduling.

8. The resource management system based on the artificial intelligence platform according to claim 7 is characterized in that: The indicator layer for predicting resource quantity changes in the computing power resource type includes the number of computing cores and accelerators, and memory capacity; the indicator layer for predicting resource quality changes in the computing power resource type includes CPU and GPU performance, and computing architecture; the indicator layer for predicting resource time periodic changes in the computing power resource type includes task response and completion time, and changes in the proportion of resource idle and busy time; the indicator layer for predicting resource quality changes in the storage resource type includes storage capacity, the number of storage devices, and the number and size of storage partitions; the indicator layer for predicting resource quality changes in the storage resource type includes read and write speed, data reliability, and storage medium life; the indicator layer for predicting resource time periodic changes in the storage resource type includes data storage cycle, data backup and recovery time, storage resource space, and usage time; the indicator layer for predicting resource quantity changes in the network resource type includes network bandwidth, the number of network ports and IP addresses; the indicator layer for predicting resource quality changes in the network resource type includes network delay, packet loss rate, and network jitter; the indicator layer for predicting resource time periodic changes in the network resource type includes network peak and trough time, data transmission time, network resource space, and busy time ratio.

9. The resource management system based on the artificial intelligence platform according to claim 8, characterized in that: The comprehensive demand weight of each resource type is calculated based on the subjective weight and the objective weight: Where: w A and are the subjective weight vector and objective weight vector of the criterion layer, and are the subjective weight vector and objective weight vector of the sub-criteria layer under the k-th resource type, and are the subjective weight vector and objective weight vector of the indicator layer under the lth sub-criteria layer under the kth resource type, respectively, and α is the combination coefficient.

10. The resource management system based on the artificial intelligence platform according to claim 9, characterized in that: The step of selecting the target demand supply end according to the comprehensive demand weight to perform resource scheduling includes: Determine input data of the hash algorithm based on the supply information fed back by the resource supply end, and construct a data structure; Determine a hash function according to the weights of the computing resource type, storage resource type, and network resource type in the resource evaluation system, and calculate the corresponding hash value; The target resource supplier is screened according to the hash value.

Images