Project demand prediction method, cloud server purchase demand prediction method and device

Abstract

The embodiment of the invention provides a project demand prediction method, a cloud server purchase demand prediction method and a cloud server purchase demand prediction device. The project demand prediction method comprises the following steps: acquiring historical time sequence data of a project demand; based on the data distribution of the historical time series data, a target project demand prediction strategy corresponding to the historical time series data is determined from at least two project demand prediction strategies, and different project demand prediction strategies correspond to the data distribution of different historical time series data; and according to a target project demand prediction strategy corresponding to the historical time sequence data, performing project demand prediction to obtain a prediction result of the project demand. The target project demand prediction strategy is obtained through data distribution and matching of the historical time series data, and project demand prediction is performed by using the matched target project demand prediction strategy to obtain the prediction result, so that the accuracy and precision of project demand prediction are ensured, and the resource utilization efficiency is improved and the cost is reduced.

Description

Technical Field

[0001] The embodiments of the present specification relate to the field of computer technology, and in particular to a project demand forecasting method, a cloud server procurement demand forecasting method and a device. Background Art

[0002] Supply chain management, from planning to execution, faces the challenge of uncertainty in every link of the supply chain, especially uncertainty on the demand side, which profoundly affects all plans and decisions in the entire supply chain.

[0003] How to minimize the impact of demand fluctuations on the quality of these decisions determines the overall cost and efficiency of the supply chain, and thus determines the competitive advantages and barriers of cloud vendors' computing services. It can be seen that in the face of demand uncertainty, the first line of defense for supply chain managers is demand forecasting. Therefore, how to use data and algorithms to accurately predict future demand fluctuations, allow the supply chain to prepare in advance, match appropriate resources, and thus achieve improved resource utilization efficiency and reduced costs is an urgent problem to be solved. Summary of the invention

[0004] In view of this, an embodiment of this specification provides a project demand forecasting method. One or more embodiments of this specification also involve a cloud server procurement demand forecasting method, a project demand forecasting device, a cloud server procurement demand forecasting device, a computing device, a computer-readable storage medium and a computer program to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a project demand forecasting method is provided, comprising:

[0006] Acquire historical time series data of project demand; based on the data distribution of the historical time series data, determine a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of historical time series data; perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data to obtain a forecast result of the project demand.

[0007] According to a second aspect of an embodiment of this specification, a cloud server procurement demand forecasting method is provided, comprising:

[0008] Obtain historical time series data of cloud server procurement demand; based on the data distribution of the historical time series data, determine a target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies, wherein different procurement demand forecasting strategies correspond to different data distributions of historical time series data; perform procurement demand forecasting according to the target procurement demand forecasting strategy corresponding to the historical time series data to obtain a forecast result of the cloud server procurement demand.

[0009] According to a third aspect of the embodiments of this specification, there is provided a project demand forecasting device, comprising:

[0010] The first acquisition module is configured to acquire historical time series data of project demand; the first determination module is configured to determine, based on the data distribution of the historical time series data, a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of historical time series data; the first forecasting module is configured to perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data, and obtain a forecast result of the project demand.

[0011] According to a fourth aspect of an embodiment of this specification, a cloud server procurement demand forecasting device is provided, including:

[0012] The second acquisition module is configured to acquire the historical time series data of the cloud server procurement demand; the second determination module is configured to determine the target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies based on the data distribution of the historical time series data, wherein different procurement demand forecasting strategies correspond to different data distributions of historical time series data; the second forecasting module is configured to perform procurement demand forecasting based on the target procurement demand forecasting strategy corresponding to the historical time series data, and obtain the forecast result of the cloud server procurement demand.

[0013] According to the fifth aspect of the embodiments of this specification, a computing device is provided, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method.

[0014] According to the sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method are implemented.

[0015] According to the seventh aspect of the embodiments of this specification, a computer program is provided, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method.

[0016] One embodiment of the present specification obtains historical time series data of project demand; based on the data distribution of historical time series data, a target project demand forecasting strategy corresponding to the historical time series data is determined from at least two project demand forecasting strategies, and different project demand forecasting strategies correspond to different data distributions of historical time series data; according to the target project demand forecasting strategy corresponding to the historical time series data, a project demand forecast is performed to obtain a forecast result of the project demand. Through the data distribution of historical time series data, the target project demand forecasting strategy corresponding to the historical time series data is matched from at least two project demand strategies, so that the matched target project demand strategy corresponds to the data distribution of the historical time series data, and the matched target project demand forecasting strategy is used to perform project demand forecasting to obtain a forecast result, thereby ensuring the accuracy and precision of the forecast of the project demand, thereby achieving an improvement in resource utilization efficiency and a reduction in cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of an interaction process under a project demand forecasting system architecture provided by an embodiment of this specification;

[0018] Figure 2 It is a framework diagram of a project demand forecasting system provided by an embodiment of this specification;

[0019] Figure 3 is a flow chart of a project demand forecasting method provided by an embodiment of this specification;

[0020] Figure 4 It is a flow chart of a cloud server procurement demand prediction method provided by an embodiment of this specification;

[0021] Figure 5a is a process flow chart of a project demand forecasting method provided by an embodiment of this specification;

[0022] Figure 5b It is a schematic diagram of project demand definition in a project demand forecasting method provided in an embodiment of this specification;

[0023] Figure 5c It is a schematic diagram of project demand modeling in a project demand forecasting method provided in an embodiment of this specification;

[0024] Figure 5d It is a schematic diagram of classification of historical time series data of project demand in a project demand forecasting method provided by an embodiment of this specification;

[0025] Figure 5e This is a schematic diagram of adjusting forecast parameters in a project demand forecasting method provided by an embodiment of this specification;

[0026] Figure 6 It is a structural schematic diagram of a project demand forecasting device provided by an embodiment of this specification;

[0027] Figure 7 It is a structural diagram of a cloud server procurement demand prediction device provided by an embodiment of this specification;

[0028] Figure 8 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0029] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0030] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0031] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0032] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0033] First, the terms involved in one or more embodiments of this specification are explained.

[0034] Cloud service: A virtualized resource service model based on the Internet, which circulates computing power as a commodity through the Internet.

[0035] Cloud server: A virtual server based on cloud computing technology. Compared with traditional physical servers, cloud servers do not rely on specific hardware devices. Instead, they use virtualization technology to run multiple virtual servers on the same physical server.

[0036] Time series forecasting: A method used to predict data values ​​at future time points. It is usually based on historical data and predicts future data values ​​by establishing mathematical models or statistical models.

[0037] Long-term stable instance demand: Cloud server instance resources that users retain for a long time, and these instances are not purchased suddenly, that is, they are not purchased in large quantities suddenly in a short period of time.

[0038] The biggest challenge facing cloud server supply chain management, from planning to execution, is the uncertainty in each link of the supply chain, especially the uncertainty on the demand side, which profoundly affects almost all decisions in the entire planning and execution chain. How to minimize the impact of demand fluctuations on the quality of these decisions determines the overall cost and efficiency of the supply chain, and thus determines the competitive advantages and barriers of cloud vendors' computing services. It can be seen that in the face of demand uncertainty, the first line of defense for supply chain managers is demand forecasting. Therefore, how to use data and algorithms to accurately predict future demand fluctuations, allow the supply chain to prepare in advance, match appropriate resources, and thus achieve improved resource utilization efficiency and reduced costs and waste is an urgent problem to be solved.

[0039] One or more embodiments of this specification focus on the long-term stable instance demand on the demand side, forecast the demand for long-term stable server instances, conduct in-depth analysis of historical long-term stable instance resource demand data, further clarify the definition of long-term stable server instance demand data and forecast the long-term stable demand data time series based on the definition. Specifically, the long-term stable instance demand instance is defined through detailed data analysis and projects, and demand forecasting is further performed for the instances selected by the above definition, and modeling problems are classified according to data distribution and multi-model forecasting solutions are designed and implemented.

[0040] Specifically, through the data distribution of historical time series data, the target project demand forecasting strategy corresponding to the historical time series data is matched from at least two project demand strategies, so that the matched target project demand strategy corresponds to the data distribution of the historical time series data, and the matched target project demand forecasting strategy is used to perform project demand forecasting and obtain forecasting results, thereby ensuring the accuracy and precision of project demand forecasting, thereby achieving improved resource utilization efficiency and reduced costs.

[0041] In this specification, a project demand forecasting method is provided. This specification also involves a cloud server procurement demand forecasting method, a project demand forecasting device, a cloud server procurement demand forecasting device, a computing device, and a computer-readable storage medium, which are described in detail one by one in the following embodiments.

[0042] See also Figure 1 , Figure 1 FIG. 1 shows a schematic diagram of an interaction process under a project demand forecasting system architecture provided by an embodiment of the present specification, such as Figure 1 As shown, the system includes a server 100 and a client 200 .

[0043] The server 100 is used to obtain historical time series data of project demand; based on the data distribution of the historical time series data, determine the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of historical time series data; perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data, and obtain a forecast result of the project demand;

[0044] Client 200: used to receive the forecast result of project demand.

[0045] By applying the solution of the embodiments of this specification, through the data distribution of historical time series data, the target project demand forecasting strategy corresponding to the historical time series data is matched from at least two project demand strategies, so that the matched target project demand strategy corresponds to the data distribution of the historical time series data, and the matched target project demand forecasting strategy is used to perform project demand forecasting and obtain forecasting results, thereby ensuring the accuracy and precision of project demand forecasting, thereby achieving improved resource utilization efficiency and reduced costs.

[0046] See also Figure 2 , Figure 2 The framework diagram of a project demand forecasting system provided by an embodiment of the present specification is shown, and the system may include a server 100 and multiple clients 200. Multiple clients 200 may establish communication connections through the server 100. In the project demand forecasting scenario, the server 100 is used to provide project demand forecasting services between multiple clients 200. Multiple clients 200 may respectively serve as senders or receivers to achieve communication through the server 100.

[0047] The user can interact with the server 100 through the client 200 to receive data sent by other clients 200, or send data to other clients 200, etc. In the project demand forecasting scenario, the user can issue a project demand forecasting request to the server 100 through the client 200, and the server 100 generates a project demand forecasting result according to the project demand forecasting request, and pushes the project demand forecasting result to other clients 200 that have established communication.

[0048] The client 200 and the server 100 are connected via a network. The network provides a medium for a communication link between the client 200 and the server 100. The network may include various connection types, such as wired or wireless communication links or optical fiber cables, etc. The data transmitted by the client 200 may need to be encoded, transcoded, compressed, etc. before being released to the server 100.

[0049] The client 200 can be a browser, an application (APP, Application), or a web application such as Hypertext Markup Language Version 5 (H5, Hyper Text Markup Language5) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The client 200 can be based on the software development kit (SDK, Software Development Kit) of the corresponding service provided by the server, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition. The client 200 can be deployed in an electronic device, and needs to rely on the device to run or some APPs in the device to run. For example, the electronic device can have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications can also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, mailbox clients, social platform software, etc.

[0050] The server 100 may include servers that provide various services, such as servers that provide communication services to multiple clients, servers for background training that provide support for models used on clients, and servers that process data sent by clients. It should be noted that the server 100 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0051] It is worth noting that the project demand forecasting method provided in the embodiments of this specification is generally executed by the server 100, but in other embodiments of this specification, the client 200 may also have similar functions as the server, thereby executing the project demand forecasting method provided in the embodiments of this specification. In other embodiments, the project demand forecasting method provided in the embodiments of this specification may also be jointly executed by the client 200 and the server 100.

[0052] See also Figure 3 , Figure 3 A flowchart of a project demand forecasting method provided by an embodiment of this specification is shown, which specifically includes the following steps.

[0053] Step 302: Obtain historical time series data of project requirements.

[0054] This specification is applied to the client and / or server of an application with project demand forecasting function. The following is an example using the server as an example.

[0055] When there is a need to predict project demand, historical time series data of the project demand will be obtained, so that the project demand can be predicted by using the historical time series data of the project demand to obtain the prediction result of the project demand.

[0056] Specifically, time series data refers to time series data, which is a data string recorded in chronological order. Time data can be a period number or a time point number. Historical time series data refers to the time series data corresponding to the time period before the current time. Historical time series data includes demand curves and supplementary data. Project demand refers to the demand for project processing, such as the procurement demand required for selling the project, the recycling demand required for leasing the project, etc. The project can be a variety of products such as cloud servers and clients. The project has the characteristics of long instance life cycle, stability, and non-abnormal scale.

[0057] There are many ways to obtain historical time series data of project requirements. The user can upload the historical time series data of project requirements to a specified location through the front end, and the server can obtain it. Alternatively, the user can upload the storage location identifier of the historical time series data of project requirements through the front end, and the server can obtain the historical time series data of project requirements based on the storage location identifier. Alternatively, the server can automatically obtain the historical time series data of project requirements from the storage area where the historical time series data is stored based on the need to predict project requirements.

[0058] The trigger condition for obtaining the historical time series data of project demand can be a start instruction for project demand forecasting set by the user. Based on the start instruction, the server starts to obtain the historical time series data of project demand to predict the project demand. The start instruction can also include a prediction cycle and a cycle of prediction results. For example, the prediction cycle can be one test per day, one test per week, etc., and the cycle of the prediction results can also be to predict daily demand data or weekly demand data.

[0059] Step 304: Based on the data distribution of the historical time series data, determine the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of the historical time series data.

[0060] Specifically, data distribution is a vivid way of describing data. Data distribution characterizes the characteristics of data distribution, such as data sparsity and length. The data distribution of historical time series data can be the sparsity of data in historical time series data, the time series length in historical time series data, etc. Project demand forecasting strategy is a strategy for demand forecasting based on project characteristics. There is a corresponding relationship between project demand forecasting strategy and data distribution. Time series data with different data distributions correspond to different project demand forecasting strategies. Project demand forecasting strategies can be statistical strategies, neural network strategies, zero value strategies, time decay weighted strategies, etc.

[0061] There are many ways to implement the method of determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data. The specific method is determined according to the actual situation and is not limited in this specification.

[0062] In a possible implementation of the present specification, based on the data distribution of historical time series data, determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies can be by obtaining the data categories of at least two project demand forecasting strategies, determining the data category to which the historical time series data belongs, and taking the project demand forecasting strategy corresponding to the data category described in the historical time series data as the target project demand forecasting strategy.

[0063] In another possible implementation of the present specification, based on the data distribution of the historical time series data, determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies includes the following steps:

[0064] Obtaining data processing attributes of at least two project demand forecasting strategies, wherein the data processing attributes represent data distribution of forecast data corresponding to the project demand forecasting strategies, and the data processing attributes are obtained by performing attribute analysis based on category attributes of data categories corresponding to the project demand forecasting strategies;

[0065] The data distribution of the historical time series data and the data processing properties of at least two project demand forecasting strategies are matched respectively to determine the target project demand forecasting strategy corresponding to the historical time series data.

[0066] Specifically, the data processing attribute refers to the data attribute of the data that can be processed by the project demand forecasting strategy. For example, the data processing attribute can be that the sparsity reaches a threshold, the time series length reaches a threshold, etc. The data processing attribute can be determined based on the category attribute of the data category, and the category attribute is extracted from the law of data distribution in the data category.

[0067] The implementation method of obtaining the data processing properties of at least two project demand forecasting strategies may be to obtain the data categories of processable data corresponding to at least two project demand forecasting strategies, extract the category attributes of the data categories, perform data analysis on the category attributes, and obtain the data processing properties of the project demand forecasting strategies.

[0068] There are many ways to respectively match the data distribution of historical time series data and the data processing properties of at least two project demand forecasting strategies, and determine the target project demand forecasting strategy corresponding to the historical time series data. The specific implementation method is determined based on actual conditions and is not limited in this specification.

[0069] In a possible implementation of the present specification, the time series length of historical time series data may be obtained, the time series length may be matched with data processing properties of at least two project demand forecasting strategies respectively, and based on the matching result, the target project demand forecasting strategy corresponding to the historical time series data may be determined.

[0070] In another possible implementation of the present specification, the sparsity and time series length of the historical time series data may be obtained, and the sparsity and time series length may be matched with the data processing properties of at least two project demand forecasting strategies respectively, and based on the matching results, the target project demand forecasting strategy corresponding to the historical time series data may be determined.

[0071] In another possible implementation of the present specification, the time series length and sparsity of historical time series data may be obtained, and the time series length and sparsity are respectively matched with data processing properties of at least two project demand forecasting strategies, and based on the matching results, the target project demand forecasting strategy corresponding to the historical time series data is determined.

[0072] By applying the solution of the embodiments of this specification, the attribute processing attributes of at least two project demand forecasting strategies obtained are matched with the data distribution of historical time series data, so that the data that can be processed by the target project demand forecasting strategy corresponding to the determined historical time series data is matched with the historical time series data, thereby achieving the adaptability of the historical time series data and the target project demand forecasting strategy, and thus improving the accuracy of demand forecasting based on historical time series data using the target project demand forecasting strategy.

[0073] In an optional embodiment of the present specification, the above steps respectively match the data distribution of the historical time series data and the data processing attributes of at least two project demand forecasting strategies to determine the target project demand forecasting strategy corresponding to the historical time series data, including the following steps:

[0074] Get the sparsity time series length of historical time series data;

[0075] Performing attribute analysis on data processing attributes of at least two project demand forecasting strategies to obtain a sparsity region threshold and a length region threshold of each project demand forecasting strategy;

[0076] The sparsity and time series length of the historical time series data are matched with the sparsity region threshold and length region threshold of at least two project demand forecasting strategies respectively to determine the target project demand forecasting strategy corresponding to the historical time series data.

[0077] Specifically, sparsity refers to the degree of sparseness of valid data contained in the data, which is usually expressed by the zero content rate. For example, if there are a lot of zero values, it can be determined that the sparsity is large. Time series length refers to the time length of the time series in the time series data. For example, if the historical time series data is the data of the past month, the time series length is one month. The sparsity region threshold refers to the threshold that constitutes the sparsity region. There are at least two sparsity region thresholds. For example, sparsity 20% to sparsity 60% constitutes a sparsity region, and sparsity 20% and sparsity 60% are both sparsity region thresholds. The length region threshold refers to the threshold that constitutes the length region. There are at least two length region thresholds. For example, length 10 to length 20 constitute a length region, and length 10 and length 20 are both length region thresholds.

[0078] The implementation method for obtaining the sparsity and time series length of historical time series data can be to perform data analysis on the historical time series data to obtain the sparsity and time series length of the historical time series data; or it can be to input the historical time series data into a data analysis model to obtain the sparsity and time series length output by the data analysis model, wherein the data analysis model is trained based on multiple sample time series data and the label sparsity and label time series length of each sample time series data.

[0079] An implementation method for performing attribute analysis on data processing attributes of at least two project demand forecasting strategies to obtain the sparsity area threshold and length area threshold of each project demand forecasting strategy may be to extract the sparsity attribute and length attribute from the data processing attributes of at least two project demand forecasting strategies to obtain the sparsity area threshold and length area threshold of each project demand forecasting strategy.

[0080] The implementation method of respectively matching the sparsity of historical time series data, the time series length, and the sparsity region threshold and the length region threshold of at least two project demand forecasting strategies to determine the target project demand forecasting strategy corresponding to the historical time series data can be to calculate the matching degree between the sparsity of the historical time series data and the sparsity region threshold of at least two project demand forecasting strategies, and select an initial project demand forecasting strategy whose matching degree meets a first matching threshold from at least two project demand forecasting strategies; calculate the matching degree between the time series length of the historical time series data and the length region threshold of the initial project demand forecasting strategy, and select a project demand forecasting strategy whose matching degree meets a second matching threshold from the initial project demand forecasting strategy as the target project demand forecasting strategy.

[0081] Optionally, it is also possible to first select an initial project demand forecasting strategy whose matching degree meets the matching threshold based on the time series length of the historical time series data, and then select a target project demand forecasting strategy whose matching degree meets the matching threshold from the initial project demand forecasting strategies based on the sparsity of the historical time series data.

[0082] By applying the scheme of the embodiments of the present specification, the sparsity and time series length of the historical time series data are obtained, and attribute analysis is performed on the data processing attributes of at least two project demand forecasting strategies to obtain the sparsity region threshold and length region threshold of each project demand forecasting strategy. The sparsity and time series length are matched based on the sparsity region threshold and the length region threshold, respectively, so that the sparsity and time series length of the target project demand forecasting strategy corresponding to the determined historical time series data can be processed by the target project demand forecasting strategy. The sparsity and time series length of the data are adapted to the sparsity and time series length of the historical time series data, thereby improving the accuracy of demand forecasting based on historical time series data using the target project demand forecasting strategy.

[0083] In an optional embodiment of the present specification, before the above step of determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data, the following steps are also included:

[0084] Determine whether the historical time series data of the project requirements meets the prediction screening conditions, which are generated by the life cycle threshold and scale threshold of the server instance corresponding to the project;

[0085] When the historical time series data of the project demand meets the forecast screening condition, a step of determining a target project demand forecast strategy corresponding to the historical time series data from at least two project demand forecast strategies based on the data distribution of the historical time series data is performed.

[0086] Specifically, the prediction screening conditions refer to the conditions for screening projects for demand forecasting. The prediction screening conditions act on historical time series data. When the historical time series data does not meet the prediction screening conditions, the method of one or more embodiments of this specification is not used to predict the demand for the project. The prediction screening conditions are generated by the life cycle threshold and scale threshold of the project strength. The scale threshold can be generated by the sudden demand threshold, the specification cluster threshold, the project type restriction, and the key customer threshold. Among them, the sudden demand threshold is that any user purchases no more than a specified number of products in one day, the specification cluster threshold is not included in the projects of the big promotion, such as the products of the shopping festival, the project type restriction is to filter except the public cloud, and the key customer threshold is to filter the projects corresponding to the customers whose project demand reaches the demand threshold.

[0087] The implementation method for determining whether the historical time series data of the project requirements meets the prediction filtering conditions can be to extract dimensional data in the historical time series data of the project requirements based on the filtering dimensions of the prediction filtering conditions; compare the dimensional data with the preset filtering conditions to determine whether the comparison result meets the conditions.

[0088] When the historical time series data of the project demand meets the prediction screening conditions, the data distribution based on the historical time series data is executed to determine the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies. The implementation method may be that when the comparison result meets the conditions, one or more embodiments of the present specification are used to predict the demand of the project to determine the target project demand forecasting strategy corresponding to the historical time series data, and the target project demand forecasting strategy is used to predict the demand of the project to obtain a prediction result.

[0089] By applying the solution of the embodiments of the present specification, preset filtering conditions are set in advance, and the preset filtering conditions of the historical time series data are compared, so that the historical time series data that is predicted meets the preset filtering conditions, and the prediction results corresponding to the predicted project needs are in line with the original intention of the solution of the embodiments of the present specification, thereby ensuring the accuracy of the obtained prediction results.

[0090] Step 306: Perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data to obtain a forecast result of the project demand.

[0091] There are many ways to perform project demand forecasting based on the target project demand forecasting strategy corresponding to the historical time series data. There are many ways to obtain the forecast results of project demand, which are determined based on actual conditions and are not limited in this specification.

[0092] In one possible implementation of the present specification, project demand forecasting is performed according to a target project demand forecasting strategy corresponding to historical time series data to obtain a forecast result of project demand. This can be achieved by building a project demand forecasting machine based on the target project demand forecasting strategy, inputting historical time series data into the project demand forecasting machine, performing project demand forecasting, and obtaining a forecast result of project demand.

[0093] In another possible implementation of this specification, according to the target project demand forecasting strategy corresponding to the historical time series data, the project demand forecast is performed to obtain the forecast result of the project demand. It is also possible to extract features from the historical time series data to obtain endogenous features and exogenous features, wherein the endogenous features represent the curve features of the demand curve corresponding to the historical time series data, and the exogenous features represent the supplementary features outside the demand curve corresponding to the historical time series data; based on the target project demand forecasting strategy, endogenous features and exogenous features, the project demand forecast is performed to obtain the forecast result of the project demand. That is, the above steps of performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data to obtain the forecast result of the project demand include the following steps:

[0094] Extract features from historical time series data to obtain endogenous features and exogenous features, where the endogenous features represent the curve features of the demand curve corresponding to the historical time series data, and the exogenous features represent the supplementary features other than the demand curve corresponding to the historical time series data;

[0095] Based on the target project demand forecasting strategy, endogenous characteristics and exogenous characteristics, project demand forecasting is performed to obtain the forecast results of project demand.

[0096] Specifically, endogenous features represent the curve features of the demand curve corresponding to the historical time series data. Curve features are features of the demand change law of the historical time series data. For example, endogenous features may be sliding window value features, lag features, etc. Exogenous features represent supplementary features other than the demand curve corresponding to the historical time series data. Exogenous features are extracted based on the supplementary data corresponding to the historical time series data. Exogenous features may be timestamp-derived features, holiday features, etc. Curve features refer to features extracted from data expressed by the curve corresponding to the historical time series data. Supplementary features refer to features extracted from supplementary data not expressed by the curve corresponding to the historical time series data.

[0097] The method for extracting features from historical time series data to obtain endogenous features and exogenous features can be to construct a demand curve based on the historical time series data, extract attribute features of the demand curve from the historical time series data based on the demand curve, use the attribute features as supplementary data, perform feature extraction based on the demand curve and the supplementary data, and obtain endogenous features and exogenous features of the historical time series data; or it can be to input the historical time series data into an endogenous feature extraction model to obtain endogenous features output by the endogenous feature extraction model, and input the historical time series data into an exogenous feature extraction model to obtain exogenous features output by the exogenous feature extraction model.

[0098] Based on the target project demand forecasting strategy, endogenous features and exogenous features, a project demand forecast is conducted to obtain the forecast results of the project demand. The implementation method is to build a forecasting machine based on the target project demand forecasting strategy, input the endogenous features and exogenous features into the forecasting machine as the data basis, and obtain the forecast results of the project demand output by the forecasting machine.

[0099] Exemplarily, the target project demand forecasting strategy is a statistical method strategy, which uses statistical methods to build a statistical model, inputs endogenous features and exogenous features into the statistical model, and obtains the forecast results of the project demand obtained by statistical forecasting using the statistical model.

[0100] Exemplarily, if the target project demand forecasting strategy is a zero-value strategy, a forecasting machine is constructed based on the zero-value, and endogenous features are input into the forecasting machine to obtain the forecasting result output by the forecasting machine. The forecasting result obtained by the zero-value strategy is usually a zero value.

[0101] Exemplarily, if the target demand forecasting strategy is a time decay weighted strategy, then an attenuation prediction machine is constructed based on the time decay weighting, and the endogenous features are input into the attenuation prediction machine to obtain the prediction result output by the attenuation prediction machine, wherein the execution logic of the attenuation prediction machine is to take the demand of the most recent point in the historical time series data and set it as x, then the demand forecast value of the first point in the future is q*x, and the demand forecast value of the second point in the future is q*q*x, wherein q is the weight value, for example, if the weight is 0.9, then the demand forecast value of the first point in the future is 0.9*x, and the demand forecast value of the second point in the future is 0.81*x.

[0102] Historical time series data is a sampling or statistics of a certain value over a period of time, so the time series is composed of multiple points, and there is a point for each time scale. For example, there are historical monthly time series demand statistics for product z from January 2023 to September 2023, so there are 9 months, which is nine points. If you want to predict the next 6 months, then you need to predict 6 points from October 2023 to March 2024.

[0103] By applying the scheme of the embodiments of the present specification, based on the data distribution of the acquired historical time series data of project demand, a corresponding target project demand forecasting strategy is determined from at least two project demand forecasting strategies, so that the strategy for project demand forecasting based on the historical time series data is the target project demand forecasting strategy corresponding to the data distribution of the historical time series data, and when forecasting the historical time series data, feature extraction is performed on the historical time series data to obtain endogenous and exogenous features, so that project demand forecasting is based on endogenous features and exogenous features, and demand forecasting is performed through the target project demand forecasting strategy, endogenous features and exogenous features corresponding to the data distribution of the historical time series data, thereby ensuring the accuracy and precision of forecasting project demand.

[0104] In an optional embodiment of the present specification, the above steps extract features from historical time series data to obtain endogenous features and exogenous features, including the following steps:

[0105] Obtaining a demand curve and supplementary data from historical time series data, wherein the supplementary data is project attribute data other than the demand curve obtained from the historical time series data;

[0106] Extract features from demand curves to obtain endogenous features of historical time series data;

[0107] Feature extraction is performed on the supplementary data to obtain exogenous features of historical time series data.

[0108] Specifically, the demand curve is obtained by intuitively expressing the demand data of historical time series data in the form of a curve. The demand curve can be a line graph with time as the horizontal axis and demand as the vertical axis. Supplementary data is project attribute data other than the demand curve obtained from historical time series data. For example, the project attribute data can be the region to which the demand curve belongs, the corresponding project, whether there have been promotional activities, and the holiday characteristics.

[0109] The method for obtaining the demand curve and supplementary data of historical time series data can be to parse the historical time series data to obtain demand data and attribute data, generate a demand curve based on the demand data, adjust the attribute data based on the demand curve, and obtain supplementary data, wherein the supplementary data corresponds to the demand curve.

[0110] The method for extracting features from the demand curve and obtaining the endogenous features of the historical time series data can be to extract the regular characteristics of the demand curve based on the demand curve and obtain the endogenous features of the historical time series data.

[0111] The method for extracting features from the supplementary data to obtain exogenous features of the historical time series data can be to extract exogenous features corresponding to the demand curve from the supplementary data based on the corresponding relationship between the supplementary data and the demand curve.

[0112] By applying the solution of the embodiments of this specification, the demand curve and supplementary data of historical time series data are obtained, and feature extraction is performed on the demand curve and the supplementary data respectively, so as to obtain the endogenous features and exogenous features of the historical time series data, so that the project demand can be predicted based on the endogenous features and exogenous features in the future, thereby ensuring the comprehensive consideration of the historical time series data when predicting the project demand, thereby improving the accuracy of the prediction results of the project demand prediction.

[0113] In an optional embodiment of the present specification, before extracting features from historical time series data, missing values ​​are filled in the historical time series data to ensure the accuracy of endogenous features and exogenous features obtained by constructing features based on the historical time series data. That is, before extracting features from the historical time series data in the above steps to obtain endogenous features and exogenous features, the following steps are also included:

[0114] Get the missing value filling strategy;

[0115] Using the missing value filling strategy, the missing values ​​of historical time series data are filled to obtain the filled historical time series data.

[0116] Specifically, the missing value filling strategy refers to a strategy for filling data with missing features. For example, the missing value filling strategy can be average value filling, zero value filling, specified value filling, etc.

[0117] The implementation method of obtaining the missing value filling strategy can be to obtain the missing value filling strategy corresponding to the missing attribute based on the missing attribute of the historical time series data; or to obtain a pre-set specified missing value filling strategy.

[0118] The missing value filling strategy is used to fill the missing values ​​of historical time series data to obtain the filled historical time series data. The implementation method can be to determine the missing positions to be filled based on the historical time series data, determine the filling values ​​of the missing positions to be filled based on the missing value filling strategy, fill the filling values ​​into the missing value positions to be filled, and obtain the filled historical time series data.

[0119] By applying the solution of the embodiment of this specification, before performing feature extraction on historical time series data, a missing value filling strategy is obtained to fill the historical time series data, so that the historical time series data for feature extraction is time series data without missing values, thereby improving the accuracy of feature extraction and further ensuring the accuracy of the prediction results for predicting project demand.

[0120] In an optional embodiment of the present specification, the project demand is a first demand or a second demand, the first demand represents a positive demand, and the second demand represents a negative demand;

[0121] The above steps obtain the historical time series data of project requirements, including the following steps:

[0122] Obtain comprehensive historical time series data;

[0123] Splitting the comprehensive historical time series data to obtain historical time series data of each of at least two project requirements, wherein the at least two project requirements include a first requirement and a second requirement;

[0124] After performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data and obtaining the forecast result of the project demand, the following steps are also included:

[0125] The forecast results of each of the at least two project requirements are integrated to obtain a comprehensive forecast result of the target requirement.

[0126] Specifically, the first demand represents positive demand. The second demand represents negative demand. For example, if the demand is equal to the creation amount minus the release amount, the first demand is the creation amount demand, and the second demand is the release amount demand. If the computing resources in the rental system are taken as projects, when an instance is created, it means that part of the computing resources corresponding to the instance are leased by the user. When the instance is released, it means that this part of the computing resources is returned by the user. If the creation amount is always greater than the release amount, it means that there is always a positive demand, which will make the idle computing resources less and less. Therefore, it is necessary to predict the demand and replenish computing resources in time to ensure that there are excess computing resources in the rental system.

[0127] For the implementation method of obtaining comprehensive historical time series data, please refer to the implementation method of obtaining historical time series data required by the project mentioned above, which will not be elaborated in this manual.

[0128] The comprehensive historical time series data can be split to obtain the historical time series data of each of at least two project requirements by extracting the historical time series data of positive project requirements and the historical time series data of negative project requirements from the comprehensive historical time series data based on the positive characteristics of the first requirement and the positive characteristics of the second requirement.

[0129] The method for integrating the prediction results of each project demand of at least two project demands to obtain the comprehensive prediction result of the target demand can be to integrate the prediction result of the first demand and the prediction result of the second demand according to the difference in positive and negative characteristics to obtain the comprehensive prediction result of the target demand; or it can be to set the prediction result of the first demand as a positive prediction result based on the positive characteristic of the first demand, and set the prediction result of the second demand as a negative prediction result based on the negative characteristic of the second demand, and add the positive prediction result to the negative prediction result to obtain the comprehensive prediction result of the target demand.

[0130] By applying the solution of the embodiments of the present specification, the comprehensive historical time series data of the target demand is split into historical time series data of at least two project demands, so that the regularity of the two historical time series data after the split is stronger, which is more conducive to the accuracy of the prediction. After respectively predicting the prediction results of at least two project demands, the two prediction results are integrated according to the characteristics of the at least two project demands, thereby ensuring the accuracy of the prediction of the target demand.

[0131] In an optional embodiment of the present specification, after the above steps perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data and obtain the forecast result of the project demand, the following steps are also included:

[0132] Obtain the benchmark time series data corresponding to the project requirements;

[0133] Compare the prediction results with the benchmark time series data to obtain the prediction difference;

[0134] When the prediction difference does not meet the difference threshold, the prediction parameters in the project demand forecasting method are adjusted based on the prediction difference, wherein the prediction parameters are parameters of at least one of the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method.

[0135] Specifically, the benchmark time series data is the labeled future time series data corresponding to the historical time series data, and the benchmark time series data is the same as the time series of the prediction result. The prediction difference refers to the difference between the true value and the predicted value. The prediction difference can be a single value or multiple differences in the form of a time series. The difference threshold is a pre-set maximum error value that can be processed for the prediction demand.

[0136] There are many ways to compare the prediction results with the reference time series data to obtain the prediction difference, which are determined according to the actual situation and are not limited in this specification.

[0137] In a possible implementation of the present specification, for a target time, the data of the target time in the prediction result is compared with the data of the target time in the reference time series data to determine a prediction sub-difference, and the prediction sub-differences of each target time are summarized to obtain a prediction difference, wherein the target time is any time in the time series of the prediction result or the reference time series data, and the method for calculating the prediction difference for a single point is detailed in the following formula (1).

[0138]

[0139] Among them, Acc t is the accuracy rate in month t, R t is the benchmark time series data of month t, F tis the prediction result, abs(x) is the absolute value of x, where x is a function auxiliary reference with no actual meaning, MAPE (Mean Absolute Percentage Error) is the mean absolute percentage error, MAPE t is the mean absolute percentage error in month t.

[0140] In another possible implementation of the present specification, the prediction results can be aligned with the reference time series data according to the time series, and the aligned prediction results and the reference time series data can be compared to obtain the prediction difference, such as calculating the multi-month weighted comprehensive difference, so as to characterize the multi-month weighted comprehensive accuracy, see the following formula (2).

[0141]

[0142] Among them, Acc (Accuracy) is the accuracy, T is the number of months, To use Acc t The calculation method is to traverse the values ​​from January to T and sum them.

[0143] Optionally, when the project demand is the first demand or the second demand, the prediction result of the first demand or the second demand is obtained, and the prediction difference is obtained based on the benchmark time series data of the first demand or the second demand and the prediction result. Since the demand for the first demand or the second demand is usually close to zero and fluctuates between positive and negative, the average accuracy is directly calculated. For details, see the following formula (3).

[0144]

[0145] When the prediction difference does not meet the difference threshold, the prediction parameters in the project demand forecasting method are adjusted based on the prediction difference. When the prediction difference does not meet the difference threshold, it can be determined that the prediction of the project demand forecasting method does not meet the prediction accuracy requirement, and then the prediction parameters in the demand forecasting method are adjusted to improve the prediction accuracy of the project demand forecasting method.

[0146] The implementation method of adjusting the prediction parameters in the project demand forecasting method based on the prediction difference may be to adjust the parameters of any one of the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method in the project demand forecasting method based on the prediction difference, return to the prediction process of executing the project demand forecasting method for the parameters of any one of the items, and adjust the step of any one of the prediction parameters based on the prediction result, and select the value of the prediction parameter with the smallest prediction difference as the benchmark prediction parameter; or it may be to traverse and combine based on the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method to obtain multiple combinations, adjust the prediction parameters in the project demand forecasting method to the next combination based on the prediction difference, return to the prediction process of executing the project demand forecasting method, and adjust the prediction parameters to a combination different from the executed combination based on the prediction result, until the prediction parameters corresponding to the combination with the smallest prediction difference are selected as the parameters of the prediction method, so as to obtain an optimized project demand forecasting method.

[0147] Apply the solution of the embodiments of this specification to obtain benchmark time series data corresponding to the project demand, compare the difference between the benchmark time series data and the prediction result, and adjust the parameters of at least one of the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method in the project demand forecasting method based on the prediction difference obtained by comparison to obtain an optimized project demand forecasting method, so that more accurate prediction results can be obtained based on the optimized project demand forecasting method in the future.

[0148] In an optional embodiment of the present specification, the above steps adjust the forecast parameters in the project demand forecasting method based on the forecast difference, including the following steps:

[0149] Based on the forecast difference, adjusting the specified forecast parameter in the project demand forecast method to obtain an updated forecast method, wherein the specified forecast parameter is a parameter of any one of the project demand forecast strategy, the method for determining the target project demand forecast strategy, and the feature extraction method;

[0150] Obtain historical time series data of project requirements that are different from the project requirements, and return to the step of executing the data distribution based on the historical time series data, determining a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, until a prediction difference that meets a difference threshold is obtained, and determining the parameter value of the specified prediction parameter in the prediction method that calculates the prediction difference as the benchmark parameter value of the specified prediction parameter.

[0151] Specifically, the specified prediction parameter is a parameter of any one of the prediction parameters, a parameter of any one of the specific location project demand prediction strategy, a method of determining the target project demand prediction strategy, and a feature extraction method.

[0152] Based on the prediction difference, the specified prediction parameters in the project demand prediction method are adjusted to obtain an updated prediction method. The implementation method can be to set an adjustment direction based on the specified prediction parameters, and adjust the specified prediction parameters in the project demand prediction method based on the adjustment direction and the prediction difference to obtain an updated prediction method.

[0153] Among them, when the project demand forecasting strategy is to specify forecasting parameters, the adjustment direction of the project demand forecasting strategy may include at least one of the loss function quantile value, the neural network parameter amount, whether to increase the equipment information sequence, whether to increase important festival features, and whether to increase or delete endogenous derivative features.

[0154] When the method for determining the target project demand forecasting strategy is to specify forecasting parameters, the adjustment direction of the method for determining the target project demand forecasting strategy may include adjusting the threshold corresponding to the determination conditions for determining the target project demand forecasting strategy and the key data for forecasting.

[0155] When the feature extraction method is to specify prediction parameters, the adjustment direction of the feature extraction may include whether to split the historical time series data before performing feature extraction.

[0156] Apply the solution of the embodiments of the present specification, adjust the specified prediction parameters in the project demand forecasting method based on the prediction difference, obtain an updated prediction method, and use the updated prediction method to predict the historical time series data of project demand that is different from the historical time series data of executed project demand to perform demand forecasting, and adjust the specified prediction parameters based on the prediction results until the minimum prediction difference corresponding to the specified prediction result is obtained, determine the parameter value of the specified prediction parameter corresponding to the minimum prediction difference as the optimized parameter value of the specified prediction parameter, use the optimized parameter values ​​to update the prediction method, and obtain an optimized prediction method, so that the optimized prediction method can be used to perform demand forecasting in the future to ensure the accuracy of forecasting project demand.

[0157] In an optional embodiment of the present specification, after the above steps perform project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data and obtain the forecast result of the project demand, the following steps are also included:

[0158] When the period to which the forecast results belong is an abnormal period, the target project demand forecast strategy is corrected based on historical time series data and anomaly correction mechanism;

[0159] Based on the revised target project demand forecasting strategy, endogenous characteristics and exogenous characteristics, project demand forecasting is performed to obtain the forecast results of project demand.

[0160] Specifically, abnormal periods represent periods when the project demand forecasting strategy is different from the actual period. For example, the characteristics of the lunar calendar of the Spring Festival are not fully compatible with the Gregorian calendar. Therefore, it may happen that the Spring Festival occurs in January, which will cause February to become an abnormal period. The project demand forecasting strategy still tends to give a low forecast value for February due to inertia, resulting in a deviation in the forecast results.

[0161] The anomaly correction mechanism refers to a mechanism that corrects the deviation of sporadic forecasts for abnormal periods in the project demand forecasting strategy. For example, the original setting was to make sporadic forecasts for the Spring Festival, but due to the incompatibility between the lunar calendar and the Gregorian calendar, the sporadic forecasts in the project demand forecasting strategy shifted to months other than the Spring Festival, and the forecast for the month actually corresponding to the Spring Festival became inaccurate. The sporadic forecasts in the project demand forecasting strategy are corrected through the anomaly correction mechanism.

[0162] The implementation method for determining whether the period to which the prediction result belongs is an abnormal period can be to obtain the prediction accuracy of the prediction result. When the prediction accuracy does not meet the accuracy threshold, the lunar calendar and Gregorian calendar time of the period to which the prediction result belongs are obtained to determine whether the lunar calendar time and the Gregorian calendar time match.

[0163] When the period to which the forecast result belongs is an abnormal period, the implementation method of correcting the target project demand forecast strategy based on historical time series data and anomaly correction mechanism can be to correct the target project demand forecast strategy based on historical time series data and anomaly correction mechanism when the lunar calendar time and the Gregorian calendar time do not match.

[0164] For example, when backtesting and verifying the most recent year, it was found that the prediction accuracy for February was very low. This problem did not occur in backtesting in previous years. Through the anomaly correction mechanism, it was found that the key factor was the drift of the Spring Festival in the natural month. The anomaly correction mechanism was used to make corrections. Specifically, when the time point to be predicted includes the Spring Festival and the Spring Festival occurs in February, no correction is required; when the Spring Festival occurs in January, the anomaly correction mechanism is used to make corrections. For example, a special model is enabled for prediction, and historical data is used to weighted fit the data values ​​of the Spring Festival period to correct the errors caused by period drift.

[0165] Based on the revised target project demand forecasting strategy, endogenous features and exogenous features, project demand forecasting is performed to obtain the forecast results of project demand. The implementation method is to use the revised target project demand forecasting strategy to build a forecasting machine, input the endogenous features and exogenous features into the forecasting machine, and obtain the forecast results of project demand output by the forecasting machine.

[0166] By applying the solution of the embodiment of this specification, it is determined whether the period corresponding to the prediction result is an abnormal period. If so, the target project demand forecasting strategy is corrected based on historical time series data and the abnormal correction mechanism, so as to use the corrected target project demand forecasting strategy to perform demand forecasting, thereby achieving the purpose of improving the accuracy of the prediction results.

[0167] In one or more embodiments of the present specification, there will be a large number of demands for project instances of different models in different regional warehouses. After the above-mentioned project demand forecasting method is used to predict any project and obtain the forecast results, the forecast results of the demands of each project are summarized to obtain all the forecast results for the project. For example, assuming there are 100 availability zones and each availability zone sells 100 types of projects, then 10,000 historical time series data will be generated. However, because the corresponding availability zones and the launch time of the projects are different, the length of the historical time series data of each project is also different. Some may be 5 years, and some may be only 3 weeks. For any of the 10,000 historical time series data, determine the corresponding target project demand forecasting strategy, among which, there are 500 historical time series data corresponding to the zero value strategy, and the prediction results corresponding to these 500 historical time series data are zero value; there are 1,000 historical time series data corresponding to the time decay weighted strategy, so the prediction is performed through the time decay weighted method to obtain the prediction results of the 1,000 historical time series data; there are 1,500 historical time series data corresponding to the statistical method strategy, and the prediction is performed using the statistical method to obtain the prediction results of the 1,500 historical time series data; there are 7,000 historical time series data corresponding to the neural network strategy, and the prediction results of the 7,000 historical time series data are obtained through neural network prediction; finally, the prediction results corresponding to 500+1,000+1,500+7,000 historical time series data are summarized to obtain the prediction results of 10,000 demand time series related to 100 availability zones and 100 products, and then the project is purchased based on the prediction results.

[0168] By applying the scheme of the embodiments of the present specification, based on the data distribution of the acquired historical time series data of project demand, a corresponding target project demand forecasting strategy is determined from at least two project demand forecasting strategies, so that the strategy for project demand forecasting based on the historical time series data is the target project demand forecasting strategy corresponding to the data distribution of the historical time series data, and when forecasting the historical time series data, feature extraction is performed on the historical time series data to obtain endogenous and exogenous features, so that project demand forecasting is based on endogenous features and exogenous features, and demand forecasting is performed through the target project demand forecasting strategy, endogenous features and exogenous features corresponding to the data distribution of the historical time series data, thereby ensuring the accuracy and precision of forecasting project demand.

[0169] See also Figure 4 , Figure 4A flowchart of a cloud server procurement demand forecasting method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0170] Step 402: Obtain historical time series data of cloud server procurement demand.

[0171] Step 404: Based on the data distribution of the historical time series data, determine a target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies, wherein different procurement demand forecasting strategies correspond to different data distributions of the historical time series data.

[0172] Step 406: Perform procurement demand forecasting according to the target procurement demand forecasting strategy corresponding to the historical time series data to obtain the forecast result of the procurement demand of the cloud server.

[0173] The above steps 402 to 406 correspond to the technical solution of the cloud server procurement demand forecasting method and the above Figure 3 The technical solutions of the corresponding project demand forecasting method belong to the same concept. For details not described in detail in the technical solutions of the cloud server procurement demand forecasting method, please refer to the description of the technical solutions of the above-mentioned project demand forecasting method.

[0174] The following combination Figure 5a Taking the application of the project demand forecasting method provided in this specification in predicting the procurement demand of cloud servers as an example, the project demand forecasting method is further explained. Figure 5a A process flow chart of a project demand forecasting method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0175] Step 502: Determine a long-term stable server instance demand definition based on demand burstiness and magnitude.

[0176] as follows Figure 5b As shown, Figure 5bA schematic diagram of project demand definition in a project demand forecasting method provided by an embodiment of the specification is shown, and a long-term stable server instance definition specifically includes long-term stable server instance data. First, the data information includes the creation amount and release amount, as well as key resource characteristics. "Creation amount, release amount and key resource characteristics" are the objects of data processing. The source data is instance-granular, including information such as the creation and release time of the instance, as well as the static information of these instances (key resource characteristics), such as the region, availability zone, and third-level product to which the instance belongs; then the information data of the user instance is recorded, and these project data are reasonably analyzed to better serve the user. In addition, the long-term stable server instance emphasizes long-term and stability by setting filtering rules. First, the long-term concept is emphasized, and the filtering threshold of the instance life cycle is adjusted to a non-long-term time (instance life cycle); secondly, the stability concept is emphasized, and the stability concept is achieved by eliminating the sudden demands of large customers and some small and medium-sized customers in certain regions / products from time to time, (business types) other than public clouds, and large-scale promotion projects (specification clusters).

[0177] Step 504: Define a modeling problem framework based on long-term stable server instance requirements.

[0178] as follows Figure 5c As shown, Figure 5cA schematic diagram of project demand modeling in a project demand forecasting method provided by an embodiment of the present specification is shown, and demand forecasting problem modeling is defined based on a long-term stable server instance, specifically including problem + indicator + feature. First, the problem modeling / prediction method is clarified. For the prediction problem, the historical time series data is split to obtain the creation volume and release volume, and the creation volume / release volume is predicted, so as to better utilize the different characteristics of the two types of curves. At the same time, because the volatility of daily demand on the time axis is relatively high, it is aggregated into weekly demand, and prediction is performed using weekly granularity data; secondly, after determining the problem, evaluation indicators are designed, using intuitive and clear trend line graphs, visually combining numerical accuracy indicators, using the creation volume / release volume to calculate the direct average accuracy, and using the demand in the prediction results to calculate the multi-month comprehensive accuracy, and the indicator evaluation is performed through trend line graphs, direct average accuracy and multi-month comprehensive accuracy. After completing the problem modeling, feature engineering is used to fill in missing values ​​and construct multiple types of endogenous and exogenous features based on the data characteristics to fully refine and express the existing information. Specifically, the missing values ​​of the original data are processed, and then the endogenous and exogenous features are constructed. The endogenous features are mainly obtained by processing the demand curve of the historical time series data of demand. The exogenous features are obtained by extracting features using other supplementary data except the demand curve. The supplementary data corresponds to the demand curve. For example, the supplementary data can be the availability zone of the demand curve, the product, whether there have been any promotional activities, holiday characteristics, etc. Among them, the constructed "timestamp derived features" and "holiday features" are exogenous features, and the "sliding window value features" and "lag features" are endogenous features.

[0179] Step 506: Design a data classification modeling prediction algorithm framework for the long-term stable server instance demand prediction problem.

[0180] as follows Figure 5d As shown, Figure 5dA schematic diagram of the classification of historical time series data of project demand in a project demand forecasting method provided by an embodiment of the present specification is shown, specifically, a data classification modeling and prediction algorithm framework in the long-term stable server instance definition demand, data analysis is performed on the overall data, and the data is classified and screened for forecasting, thereby avoiding mutual interference between data of different types during forecasting, and at the same time improving the prediction accuracy of each type of data by utilizing the unique characteristics of each type of data. Specifically, the data is divided according to the sparsity and length, and the mechanism strategy, statistical trend prediction method and neural network method are used for prediction for different data. For example, the data is classified and it is identified that the historical time series data is "more zero values ​​+ short length", then the zero value strategy is used for demand prediction; when the historical time series data is identified as "extremely many zero values", the time decay weighted strategy is used for demand prediction; when the historical time series data is identified as "relatively normal distribution + normal length", the TFT neural network strategy is used to make full use of the feature information for demand prediction; when the historical time series data is identified as "relatively normal distribution + shorter length", the PROPHET statistical method strategy is used, which uses less feature information to predict the demand. If there are multiple historical time series data and each category has corresponding historical time series data, after obtaining the corresponding prediction results of each category, the prediction results of each category are summarized to obtain the full prediction results. Among them, TFT (Temporal Fusion Transformers) is a deep learning model that can explain the time series prediction results, and the PROPHET statistical method can form a decomposable time series model.

[0181] Step 508: Design a multi-dimensional experiment of model features based on the data classification modeling prediction algorithm framework.

[0182] as follows Figure 5e As shown, Figure 5e A schematic diagram of adjustment of prediction parameters in a project demand forecasting method provided by an embodiment of the present specification is shown, specifically a multi-dimensional test of model features in the long-term stable server instance definition demand, that is, based on the aforementioned basic prediction algorithm framework, a control experiment (baseline experiment) is conducted, specifically for independent or combined prediction methods (split into create / release for separate predictions, or aggregating historical time series data to directly predict demand), for whether the prediction results are subject to boundary restrictions (peak and valley values ​​or mean values), for the prediction model's own loss function and parameter quantity settings (loss function quantile values, neural network parameter quantities), for data distribution of data classification (adjusting the category thresholds of each data category, and focusing on prediction of key data), for the use of data features (adding equipment information sequences, adding important holiday features, adding and deleting endogenous derived features), and other different dimensions of the prediction level. At the same time, comparative experiments are conducted on the data distribution and data features on the data side, so as to select improvement plans that effectively improve the basic model.

[0183] Step 510: Design an anomaly correction mechanism to optimize multi-dimensional experimental results.

[0184] Specifically, introducing Spring Festival-related features into the algorithm framework (for example, the number of days that the natural week intersects with the week before and after the Spring Festival) can significantly enhance the model's learning and expression capabilities. However, because the characteristics of the Spring Festival lunar calendar are not fully compatible with the Gregorian calendar, it may happen that when the Spring Festival occurs in January, the model will still tend to focus on the low forecast value for February out of inertia. Therefore, this manual designs a mechanism to use historical months to correct the model's inertial low forecast for February when the Spring Festival occurs in January, thereby completing the closed loop of the algorithm strategy to handle the demand forecasting problem.

[0185] Step 512: Deploy the model of the time series forecasting framework and continuously monitor and optimize the forecasting results.

[0186] The time series forecasting framework is scheduled with daily granularity. The latest project data is pulled through the open data processing service every day, and input into the forecasting framework for feature engineering, historical time series data classification and screening, model loading, model forecasting and other steps to produce forecast results for all project requirements. The final forecast results will be continuously monitored and analyzed, and sporadic phenomena will be automatically judged and the abnormal correction mechanism will be called for correction, realizing a closed-loop process for model optimization.

[0187] In one or more embodiments of the present specification, careful consideration is first given to the definition of demand timing data, and filtering rules are set according to data information such as key resource characteristics and based on factors such as server life cycle / project type / burst demand / customer category to obtain long-term stable server instance data for prediction.

[0188] After that, we further modeled the demand forecasting problem for long-term stable server instances. The forecasting method was designed to split multiple curves for separate forecasting. The evaluation indicators were comprehensively evaluated from multiple aspects such as visual trend lines and numerical indicators (for example, direct average accuracy / multi-month comprehensive accuracy). In data feature engineering, data repair such as missing value filling was performed, and a large number of time-series derived endogenous and exogenous features were constructed for data enhancement. Ultimately, the model was guided from multiple angles to better fit complex data and conduct a detailed evaluation.

[0189] When designing the algorithm framework after problem modeling, this manual comprehensively considers and combines a variety of different types of time series prediction methods (for example, deep learning neural network type / traditional machine learning statistical method type / mechanism strategy type) and takes advantage of their strengths and avoids their weaknesses, and gives full play to the advantages of the corresponding time series prediction methods for data with different distributions and characteristics, thereby avoiding overfitting or underfitting of a single model. On this basis, through multi-dimensional experiments on model features, experiments are further conducted on different dimensions of the prediction level, such as whether the prediction method is independent or combined, whether to impose secondary boundary restrictions on the prediction results, and the loss function and parameter setting of the prediction model itself, on the basic framework. At the same time, comparative experiments are conducted on the data distribution and data characteristics on the data side, so as to select improvement plans that effectively improve the basic model.

[0190] Finally, in response to the occasional phenomenon caused by the incomplete adaptation of the characteristics of the lunar calendar for the Spring Festival and the Gregorian calendar, that is, when the Spring Festival occurs in February, the model still tends to focus on January out of inertia. This manual designs an abnormal correction mechanism to use historical months to correct the inertia phenomenon of the model when the Spring Festival occurs in February, further improve the performance of the algorithm framework, and finally complete the closed loop of the algorithm strategy to handle the demand forecasting problem.

[0191] In summary, the multi-model time prediction algorithm framework proposed in one or more embodiments of this specification has achieved remarkable results in the long-term stable server instance demand prediction tasks with complex project scenarios.

[0192] Corresponding to the above method embodiment, this specification also provides a project demand forecasting device embodiment, Figure 6 FIG. 1 is a schematic diagram showing the structure of a project demand forecasting device provided by an embodiment of the present specification. Figure 6 As shown, the device comprises:

[0193] A first acquisition module 602 is configured to acquire historical time series data of project requirements;

[0194] The first determination module 604 is configured to determine the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data, wherein different project demand forecasting strategies correspond to different data distributions of the historical time series data;

[0195] The first prediction module 606 is configured to perform project demand prediction according to the target project demand prediction strategy corresponding to the historical time series data to obtain the prediction result of the project demand.

[0196] Optionally, the first determination module 604 is further configured to obtain data processing attributes of at least two project demand forecasting strategies, wherein the data processing attributes characterize the data distribution of the forecast data corresponding to the project demand forecasting strategy, and the data processing attributes are obtained by performing attribute analysis based on the category attributes of the data category corresponding to the project demand forecasting strategy; respectively match the data distribution of the historical time series data and the data processing attributes of at least two project demand forecasting strategies to determine the target project demand forecasting strategy corresponding to the historical time series data.

[0197] Optionally, the first determination module 604 is further configured to obtain the sparsity and time series length of historical time series data; perform attribute analysis on the data processing attributes of at least two project demand forecasting strategies to obtain the sparsity area threshold and length area threshold of each project demand forecasting strategy; respectively match the sparsity of the historical time series data, the time series length and the sparsity area threshold and length area threshold of at least two project demand forecasting strategies to determine the target project demand forecasting strategy corresponding to the historical time series data.

[0198] Optionally, the first prediction module 606 is further configured to perform feature extraction on the historical time series data to obtain endogenous features and exogenous features, wherein the endogenous features represent the curve features of the demand curve corresponding to the historical time series data, and the exogenous features represent the supplementary features outside the demand curve corresponding to the historical time series data; based on the target project demand forecasting strategy, endogenous features and exogenous features, perform project demand forecasting to obtain the forecast results of the project demand.

[0199] Optionally, the first prediction module 606 is further configured to obtain a demand curve and supplementary data of historical time series data, wherein the supplementary data is project attribute data other than the demand curve obtained from the historical time series data; perform feature extraction on the demand curve to obtain endogenous features of the historical time series data; perform feature extraction on the supplementary data to obtain exogenous features of the historical time series data.

[0200] Optionally, the project demand is a first demand or a second demand, the first demand represents a positive demand, and the second demand represents a negative demand; the first acquisition module 602 is further configured to obtain comprehensive historical time series data; the comprehensive historical time series data is split to obtain historical time series data of each project demand in at least two project demands, wherein the at least two project demands include the first demand and the second demand; the project demand forecasting device also includes an integration module, which is configured to integrate the forecast results of each project demand in at least two project demands to obtain a comprehensive forecast result of the target demand.

[0201] Optionally, the project demand forecasting device also includes a screening module, which is configured to determine whether the historical time series data of the project demand meets the forecast screening conditions, and the forecast screening conditions are generated by the life cycle threshold and scale threshold of the server instance corresponding to the project; when the historical time series data of the project demand meets the forecast screening conditions, a step of determining a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data is executed.

[0202] Optionally, the project demand forecasting device also includes an adjustment module, which is configured to obtain benchmark time series data corresponding to the project demand; compare the forecast results with the benchmark time series data to obtain a forecast difference; when the forecast difference does not meet the difference threshold, adjust the forecast parameters in the project demand forecasting method based on the forecast difference, wherein the forecast parameters are parameters of at least one of the project demand forecasting strategy, the method of determining the target project demand forecasting strategy, and the feature extraction method.

[0203] Optionally, the project demand forecasting device also includes an adjustment module, which is configured to adjust specified forecasting parameters in the project demand forecasting method based on the prediction difference to obtain an updated forecasting method, wherein the specified forecasting parameters are parameters of any one of the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method; obtain historical time series data of project demand different from the project demand, and return to execute the step of determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data until a prediction difference that meets the difference threshold is obtained, and the parameter value of the specified prediction parameter in the prediction method for calculating the prediction difference is determined as the benchmark parameter value of the specified prediction parameter.

[0204] Optionally, the project demand forecasting device also includes a correction module, which is configured to correct the target project demand forecasting strategy based on historical time series data and an abnormal correction mechanism when the period to which the forecast result belongs is an abnormal period; perform project demand forecasting based on the corrected target project demand forecasting strategy to obtain a forecast result of the project demand.

[0205] By applying the solution of the embodiments of this specification, through the data distribution of historical time series data, the target project demand forecasting strategy corresponding to the historical time series data is matched from at least two project demand strategies, so that the matched target project demand strategy corresponds to the data distribution of the historical time series data, and the matched target project demand forecasting strategy is used to perform project demand forecasting and obtain forecasting results, thereby ensuring the accuracy and precision of project demand forecasting, thereby achieving improved resource utilization efficiency and reduced costs.

[0206] The above is a schematic scheme of a project demand forecasting device of this embodiment. It should be noted that the technical scheme of the project demand forecasting device and the technical scheme of the project demand forecasting method described above are of the same concept. For details not described in detail in the technical scheme of the project demand forecasting device, please refer to the description of the technical scheme of the project demand forecasting method described above.

[0207] Corresponding to the above method embodiment, this specification also provides a cloud server procurement demand forecasting device embodiment, Figure 7 FIG. 1 is a schematic diagram showing a structure of a cloud server procurement demand forecasting device provided by an embodiment of the present specification. Figure 7 As shown, the device comprises:

[0208] The second acquisition module 702 is configured to acquire historical time series data of cloud server procurement demand;

[0209] The second determination module 704 is configured to determine, based on the data distribution of the historical time series data, a target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies, wherein different procurement demand forecasting strategies correspond to different data distributions of the historical time series data;

[0210] The second prediction module 706 is configured to perform procurement demand prediction according to the target procurement demand prediction strategy corresponding to the historical time series data, and obtain the prediction result of the procurement demand of the cloud server.

[0211] The above is a schematic scheme of a cloud server procurement demand forecasting device of this embodiment. It should be noted that the technical scheme of the cloud server procurement demand forecasting device and the technical scheme of the above-mentioned cloud server procurement demand forecasting method belong to the same concept, and the details not described in detail in the technical scheme of the cloud server procurement demand forecasting device can be found in the description of the technical scheme of the above-mentioned cloud server procurement demand forecasting method.

[0212] Figure 8 The structure block diagram of a computing device provided by an embodiment of the present specification is shown. The components of the computing device 800 include but are not limited to a memory 810 and a processor 820. The processor 820 is connected to the memory 810 via a bus 830, and the database 850 is used to store data.

[0213] The computing device 800 also includes an access device 840 that enables the computing device 800 to communicate via one or more networks 860. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 540 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0214] In one embodiment of the present specification, the above components of the computing device 800 and Figure 8 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 8 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0215] The computing device 800 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 800 may also be a mobile or stationary server.

[0216] Among them, the processor 820 is used to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method.

[0217] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method are of the same concept. For details not described in detail in the technical scheme of the computing device, please refer to the description of the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method.

[0218] An embodiment of the present specification also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method.

[0219] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method are of the same concept. For details not described in detail in the technical scheme of the storage medium, please refer to the description of the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method.

[0220] An embodiment of the present specification also provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned project demand forecasting method and cloud server procurement demand forecasting method.

[0221] The above is a schematic scheme of a computer program of this embodiment. It should be noted that the technical scheme of the computer program and the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method are of the same concept. For details not described in detail in the technical scheme of the computer program, please refer to the description of the technical scheme of the project demand forecasting method and the cloud server procurement demand forecasting method.

[0222] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0223] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0224] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0225] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0226] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A project demand forecasting method, comprising: Obtain historical time series data of project requirements; Based on the data distribution of the historical time series data, determining a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of the historical time series data; According to the target project demand forecasting strategy corresponding to the historical time series data, project demand forecasting is performed to obtain the forecast result of the project demand.

2. According to the method of claim 1, the step of determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data comprises: Acquire data processing attributes of at least two project demand forecasting strategies, wherein the data processing attributes represent data distribution of forecast data corresponding to the project demand forecasting strategies, and the data processing attributes are obtained by performing attribute analysis based on category attributes of data categories corresponding to the project demand forecasting strategies; The data distribution of the historical time series data and the data processing properties of the at least two project demand forecasting strategies are matched respectively to determine the target project demand forecasting strategy corresponding to the historical time series data.

3. According to the method of claim 2, the step of respectively matching the data distribution of the historical time series data and the data processing properties of the at least two project demand forecasting strategies to determine the target project demand forecasting strategy corresponding to the historical time series data comprises: Obtaining the sparsity and time series length of the historical time series data; Performing attribute analysis on the data processing attributes of the at least two project demand forecasting strategies to obtain a sparsity region threshold and a length region threshold of each project demand forecasting strategy; The sparsity of the historical time series data, the time series length, and the sparsity region threshold and the length region threshold of the at least two project demand forecasting strategies are matched respectively to determine the target project demand forecasting strategy corresponding to the historical time series data.

4. According to the method of claim 1, the step of performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data to obtain the forecast result of the project demand comprises: Performing feature extraction on the historical time series data to obtain endogenous features and exogenous features, wherein the endogenous features represent curve features of the demand curve corresponding to the historical time series data, and the exogenous features represent supplementary features other than the demand curve corresponding to the historical time series data; Based on the target project demand forecasting strategy, the endogenous characteristics and the exogenous characteristics, project demand forecasting is performed to obtain a forecast result of the project demand.

5. According to the method of claim 4, the step of extracting features from the historical time series data to obtain endogenous features and exogenous features comprises: Acquire a demand curve and supplementary data of the historical time series data, wherein the supplementary data is item attribute data other than the demand curve acquired from the historical time series data; Extracting features from the demand curve to obtain endogenous features of the historical time series data; Feature extraction is performed on the supplementary data to obtain exogenous features of the historical time series data.

6. The method according to claim 1, wherein the project demand is a first demand or a second demand, the first demand represents a positive demand, and the second demand represents a negative demand; The obtaining of historical time series data of project requirements includes: Obtain comprehensive historical time series data; Splitting the comprehensive historical time series data to obtain historical time series data of each of at least two project requirements, wherein the at least two project requirements include a first requirement and a second requirement; After performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data and obtaining the forecast result of the project demand, the method further includes: The forecast results of each of the at least two project requirements are integrated to obtain a comprehensive forecast result of the target requirement.

7. The method according to claim 1, before determining the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies based on the data distribution of the historical time series data, further comprises: Determine whether the historical time series data of the project demand meets the prediction screening condition, where the prediction screening condition is generated by the life cycle threshold and scale threshold of the server instance corresponding to the project; When the historical time series data of the project demand meets the prediction screening conditions, the step of determining the target project demand prediction strategy corresponding to the historical time series data from at least two project demand prediction strategies based on the data distribution of the historical time series data is performed.

8. The method according to any one of claims 1 to 7, after performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data to obtain the forecast result of the project demand, further comprising: Obtaining benchmark time series data corresponding to the project requirements; Comparing the prediction result with the reference time series data to obtain a prediction difference; When the prediction difference does not meet the difference threshold, the prediction parameters in the project demand forecasting method are adjusted based on the prediction difference, wherein the prediction parameters are parameters of at least one of the project demand forecasting strategy, the method for determining the target project demand forecasting strategy, and the feature extraction method.

9. The method according to claim 8, wherein adjusting the forecast parameters in the project demand forecasting method based on the forecast difference comprises: Based on the prediction difference, adjusting the specified prediction parameters in the project demand prediction method to obtain an updated prediction method, wherein the specified prediction parameters are parameters of any one of the project demand prediction strategy, the method for determining the target project demand prediction strategy, and the feature extraction method; Acquire historical time series data of project requirements that are different from the project requirements, and return to execute the data distribution based on the historical time series data, determine the target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, until a prediction difference that meets the difference threshold is obtained, and determine the parameter value of the specified prediction parameter in the prediction method that calculates the prediction difference as the benchmark parameter value of the specified prediction parameter.

10. The method according to claim 1, after performing project demand forecasting according to the target project demand forecasting strategy corresponding to the historical time series data and obtaining the forecast result of the project demand, further comprises: In the case that the period to which the forecast result belongs is an abnormal period, based on the historical time series data and the abnormal correction mechanism, the target project demand forecast strategy is corrected; Based on the revised target project demand forecasting strategy, project demand forecasting is performed to obtain the forecast results of project demand.

11. A method for predicting cloud server procurement demand, comprising: Obtain historical time series data on cloud server procurement demand; Based on the data distribution of the historical time series data, determining a target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies, wherein different procurement demand forecasting strategies correspond to different data distributions of the historical time series data; Procurement demand forecasting is performed according to the target procurement demand forecasting strategy corresponding to the historical time series data to obtain the forecast result of the procurement demand of the cloud server.

12. A project demand forecasting device, comprising: A first acquisition module is configured to acquire historical time series data of project requirements; A first determination module is configured to determine, based on the data distribution of the historical time series data, a target project demand forecasting strategy corresponding to the historical time series data from at least two project demand forecasting strategies, wherein different project demand forecasting strategies correspond to different data distributions of the historical time series data; The first prediction module is configured to perform project demand prediction according to the target project demand prediction strategy corresponding to the historical time series data to obtain the prediction result of the project demand.

13. A cloud server procurement demand forecasting device, comprising: A second acquisition module is configured to acquire historical time series data of cloud server procurement demand; A second determination module is configured to determine, based on the data distribution of the historical time series data, a target procurement demand forecasting strategy corresponding to the historical time series data from at least two procurement demand forecasting strategies, wherein different procurement demand forecasting strategies correspond to different data distributions of the historical time series data; The second prediction module is configured to perform procurement demand prediction according to the target procurement demand prediction strategy corresponding to the historical time series data, and obtain the prediction result of the procurement demand of the cloud server.

14. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 11 are implemented.

15. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 11.

Images